Fear before a major decision is not a signal to stop; it is a measuring instrument. The stronger the anxiety about a particular choice, the higher the cost of a possible mistake—and that is exactly why the decision deserves to be taken seriously and weighed more carefully than a routine purchase. But a leader who confuses two different things—“taking it seriously” and “dragging it out endlessly”—turns a useful signal into a brake in perfect working order. The rest of this chapter tells the story of one month in which a diversified industrial holding company nearly lost a deal that could have doubled one of its divisions; it diagnoses what exactly got stuck in the mind of the executive leading that deal; and it sets out a protocol that forces you through the phase of doubt quickly—instead of getting stuck in it for good.

The silence that cost weeks

Before turning to the story, it’s worth anchoring the scale of the problem in numbers—not invented ones, but real figures gathered from a large sample of executives. An international survey by Oracle, conducted in 2023 among more than 14,000 respondents in seventeen countries, found that 85% of business leaders had experienced what the researchers called “decision distress”—regret, guilt or lingering doubt about a decision made in the past year. A separate 2024 survey by HSBC and the research firm PSB Insights, with a sample of 17,555 respondents, recorded a starker figure: 28% of executives openly admit that uncertainty literally paralyzes them—not slows them down, but stops them in their tracks. At the same time, it’s important not to confuse slowness with caution: a McKinsey study of about 1,200 managers found that organizations where decisions are made quickly are almost twice as likely to show high decision quality as well. Speed and quality of decision-making are not opposites, as the intuitive managerial logic of “measure twice, cut once” would have it. More often they go hand in hand, and delay in itself almost never buys a better decision—it only buys a postponement.

There is also a scientific caveat worth making right away, so as not to slide into oversimplification. The popular idea of the “paradox of choice”—the more options a person faces, the more paralyzed they become—does not hold up as a universal law. A large meta-analysis by the psychologist Benjamin Scheibehenne and colleagues (2010, around fifty studies) showed that the average choice-overload effect across the entire body of data is close to zero. Decision paralysis arises not from the number of options as such but selectively—when high uncertainty, a high cost of error and low comparability between the alternatives come together. That is precisely the set of conditions typical of a large-scale strategic decision—and precisely the situation in which one particular CEO in our story found himself.

Picture a holding company—let’s call it Northern Perimeter, since this is a composite, illustrative case rather than a specific existing company. Three divisions: metalworking, logistics with its own vehicle fleet, and energy services for industrial facilities. The group’s total revenue is in the range of ₽40–45 billion a year, its headcount about 6 thousand people, its headquarters in a regional capital, and its production sites in four regions. An ordinary, stable diversified holding company—not the fastest-growing, but not a troubled one either; industry has dozens like it.

A proposal landed on the desk of Northern Perimeter’s CEO—let’s call him Igor Valentinovich: buy a regional competitor in the energy services segment. The asset was debt-laden but held the number two position in the regional market, had long-term contracts with two large industrial customers, and ran worn but serviceable equipment. The price was a third below market by the sector’s multiples, because the seller was a bank that had taken the asset over for unpaid debts and had no wish to keep it on its balance sheet. The exclusivity period for negotiations was 45 days. After that, the bank would put the asset up for an open tender, in which two larger national players had already confirmed their interest.

On paper, the deal looked obvious. The financial model prepared by the strategic development team showed a return on invested capital higher than that of any of the holding company’s organic projects over the previous three years. Legal due diligence found no critical risks—only the usual list of comments on lease agreements and environmental orders, all of them fixable. The board had given its preliminary go-ahead for negotiations back when the price was announced.

And then—silence. For thirty of the forty-five days allotted, Igor Valentinovich would not let the negotiating team move beyond the indicative offer. He didn’t reject the deal outright—he kept finding something to double-check. He ordered a second independent valuation of the asset, although the first had already been independent. He asked for the scenario with a 15% drop in rates to be recalculated once more, although that scenario was already in the model. He called an additional meeting with the lawyers on an issue they had closed two weeks earlier. Each of these steps looked like reasonable caution. Taken together, they meant one thing: the decision was not being made.

By the end of the third week, the head of the energy services division, who had led the deal from the very beginning, no longer understood what was going on. The group’s CFO said it plainly at one of the meetings: “If we’re not moving, we’re losing time, and time costs money, because the exclusivity window won’t stretch forever.” Igor Valentinovich agreed—and the next day once again asked to “double-check everything one more time.”

What actually got stuck

If we look at this story through the emotional cycle—the very “Search → Decision → Action → Reward” cycle examined in detail in Part I of the book—we can see exactly where the bottleneck occurred.

Let’s briefly recall the mechanics. Any encounter with a new task triggers the Search phase: the brain assesses the situation—whether it matters, whether there are resources to cope with it, what a mistake would cost. A key role here is played by the amygdala, a small structure deep in the brain that the neuroscientist Joseph LeDoux described in detail as a fast threat detector that works before any conscious reflection. It is the amygdala that generates the initial fear—not as a weakness but as an extremely fast filter: “this situation could cost a lot; pay attention.” The psychologist Richard Lazarus, author of the cognitive appraisal theory of emotion, put it more precisely: an emotion is not a malfunction of thinking but the result of assessing a situation on several parameters at once—how significant it is for a person’s goals and whether they have enough resources to deal with it. Fear is the answer to the second question specifically, when there is no confidence in one’s resources yet.

Igor Valentinovich’s problem was not that he felt fear in the face of a deal of this scale. The problem was that he used the fear for the wrong purpose. Working with fear properly at the entrance to the Search phase is a one-time, quick calibration: “the decision is big, a mistake is expensive, so it’s worth spending proportionally more time and resources on checking than on a routine purchase.” That would have taken the team a few days. Instead, the fear became grounds for never finishing the Search phase at all—for returning to its beginning again and again under the pretext of yet another round of checks.

A second mechanism is at work here, one described by the psychologist Jeffrey Gray in his revised theory of behavioral inhibition: at signs of risk, a system kicks in physiologically that slows a person’s movement toward the goal. Sometimes that is useful, because it holds them back from a rash step; but at a persistently high level of anxiety, this system can keep a person in a “don’t move” state for much longer than a realistic assessment of the risk requires. Ordering what was by then the third independent valuation added no new information—legally and financially, the question had been closed after the first two. It added only a postponement of the moment when he would have to say “yes” or “no” and take responsibility for the consequences.

What was happening to Igor Valentinovich himself within the holding company deserves a note of its own. Three years earlier he had already led an M&A deal: the holding company bought a logistics asset that turned out to be burdened with hidden liabilities and took two years to straighten out at the expense of the rest of the group. The board at the time openly raised questions about the quality of the pre-acquisition due diligence. Formally, no conclusions were drawn from that episode, and no penalties were imposed. Informally, the cost of a mistake in a deal like this was, for Igor Valentinovich, not abstract but lived through. That is valuable information for the assessment—and at the same time a source of excessive caution: the amygdala does not distinguish between “this particular deal is different from that one” and “deals of this type are dangerous in general”; it reacts to the category of situation as a whole if that category was once linked to a painful outcome.

A telling detail: Charles Carver and Michael Scheier, specialists in behavioral self-regulation, described what happens to a person when a goal becomes hard to reach or too risky—so-called “goal disengagement,” a state close to standstill and dejection that usually precedes setting a new, more realistic goal. What made the situation at Northern Perimeter unusual was that Igor Valentinovich did not disengage from the goal explicitly—he didn’t say “this deal isn’t for us,” and he didn’t switch to another task. He got stuck in an in-between state: formally the goal remained in force, but action toward it had in fact stopped. This is worse than an honest refusal—because a refusal at least frees up the team’s resources for something else. The thirty days of silence kept in limbo not only the deal but the whole negotiating team, which could neither see the matter through nor move on to the next task.

A separate word is due on the pressure from the board and the market. While Igor Valentinovich was requesting a fourth round of checks, the CFO noted that the national competitors had already commissioned their own legal due diligence on the asset—meaning they were seriously considering taking part in the open tender once exclusivity expired. The outside world created additional, entirely rational time pressure, but inside the deal team that pressure was not translated into action—it only added anxiety to already excessive anxiety, reinforcing behavioral inhibition even further instead of releasing it.

There is one more subtlety that is easy to miss if you view the story only as the individual weakness of one executive. The thirty days of silence were not just him getting stuck personally—the entire management machine around him was stuck. The CFO saw the problem but raised it only at meetings, never in a separate written escalation with a clear date by which an answer was needed. The division head, who understood the cost of delay better than anyone, had no authority to take the issue directly to the board over the CEO’s head—and under the company’s unwritten rules, such a move would have been seen as something close to mutiny. In other words, Igor Valentinovich would not have stayed stuck in the anxious part of the Search phase for so long if there had been a structural mechanism around him requiring that a delay on a decision of a certain size be escalated after a fixed number of days. That mechanism did not exist—it was precisely what the protocol described below created, and that is why the protocol proved valuable not just for one deal but for all of the company’s subsequent practice.

It is also useful to separate one emotion from another, even though outwardly they looked alike. Part of what could have been mistaken for growing anxiety about the deal itself was in fact the team’s irritation at the uncertainty of the process—and that, counterintuitively, is a different emotional state with a different physiology. The psychologist Eddie Harmon-Jones and colleagues showed experimentally that anger and the arousal associated with it, despite their negative coloring, belong physiologically not to the avoidance system but to the system of approach toward a goal—the very same system responsible for enthusiasm and productive tension. The irritation of the CFO and the division head was in this sense a healthy signal: it meant a readiness to move that was running up against someone else’s inhibition. Mistaking this irritation for general anxiety around the deal and trying to “calm everyone down” would have been a managerial mistake—the right response was not to suppress the irritation but to channel it into a structure that makes movement possible.

It’s also worth keeping in mind the cost of delay in its pure form—not as a metaphor but as a documented fact from an entirely different industry. In 2021 the container ship Ever Given blocked the Suez Canal for six days. By the estimate of the maritime publication Lloyd’s List, each day of standstill held up about $9.6 billion in cargo—some $55–60 billion over six days. And not because decisions about the ships’ routes were wrong, but because the canal was physically blocked and everyone in the queue had to wait. This is an extreme, vivid example of a general rule of the networked economy: the cost of inaction is not linear; it often grows faster than it seems to someone waiting for “a slightly more complete picture.” Exactly the same thing—only at a slower, less noticeable pace—was happening with each day of silence around the Northern Perimeter deal: the exclusivity window won’t stretch, and competitors don’t wait.

A protocol that forces you through the Search phase

The turning point came not from an epiphany but from structure. On the thirty-first day, two weeks before exclusivity ran out, the chairman of the board—an engineer by training, not inclined to psychological explanations—said bluntly at a meeting: “We are no longer discussing whether to buy the asset or not. We are discussing how long it will take us to decide, and by what rule.” That wording shifted the conversation from “is the deal scary or not” to “how do we organize the process of deciding on it.” This shift is the very essence of the protocol, which is worth breaking down step by step—because it can be reproduced for any major decision, not just this particular deal.

Before moving on to the steps, it is useful to recall where the idea of “deciding quickly on incomplete data” comes from in serious management practice, rather than in the intuition of one particular executive. The Nobel laureate in economics Herbert Simon showed back in the mid-twentieth century that under a real shortage of time and information, a manager does not maximize but “satisfices”—settles for the first acceptable solution that meets a threshold of requirements—and that this is rational, not a sign of weakness or insufficient thoroughness. The psychologist Gary Klein, having studied dozens of experienced fire commanders making decisions at real fires, showed that in the overwhelming majority of cases—over 80%—an experienced person does not compare options against one another but immediately recognizes the situation as a familiar pattern and acts on accumulated experience, only mentally testing the chosen option for soundness. And the entrepreneur Jeff Bezos gave a thoroughly practical formulation of the same principle in his 2016 letter to Amazon shareholders: “Most decisions should probably be made with somewhere around 70% of the information you wish you had. If you wait for 90%, in most cases, you’re probably being slow.” All three ideas—from different decades, different fields and tasks of different scale—converge on one point: a quality decision does not require exhaustive information; it requires sufficient information and a clear rule for when it is sufficient.

Step one. Recognize fear as a legitimate starting point, not a cause for shame. At the meeting it was said out loud: anxiety before a deal of this scale is not a weakness on the CEO’s part but a normal, physiologically explicable reaction to a decision with a high cost of error. The chairman’s words were almost literally these: “If you weren’t anxious about the amount in this contract, I’d be anxious about you.” This step sounds like a trifle, but it is precisely what removes the main side effect of unacknowledged fear—the need to disguise it as rational arguments (“let’s double-check once more”) instead of discussing it directly. As long as anxiety goes unnamed, it runs the process from the shadows; once named, it becomes a subject of discussion that can be bounded by a deadline.

Step two. Turn fear into a checklist of specific questions—and close each item once. Instead of an endless “let’s double-check once more,” the team drew up a closed list: what exactly must be true for the deal to be justified, and what exactly must be false for them to walk away from it. The list had six items: confirming the two key contracts with industrial customers directly with the customers themselves, not just from the bank’s documents; an independent technical assessment of the equipment’s condition with an estimate of its remaining service life; recalculating the debt load after integration into the group; a legal opinion on the environmental orders with a fixed amount of possible fines; a meeting with the key engineering staff of the asset being acquired to gauge the risk of mass departures after the deal; and a scenario calculation for a 15% and a 25% drop in rates. Each item was assigned an owner and a deadline—not “as soon as possible” but a calendar date. Crucially, once an item is closed, it is not revisited without new factual information. This rule strikes directly at the mechanism that had kept the deal in limbo: previously, what triggered a new round of checks was not the appearance of new facts but persistent anxiety.

The same principle is captured in crisis management by the “premortem” technique, popularized by the same Gary Klein: before starting a project, the team imagines in advance that it has failed and lists the likely causes of failure. This structures anxiety into a finite list of verifiable risks, instead of leaving it as a vague background feeling that one can return to endlessly. Northern Perimeter’s checklist was, in essence, a premortem after the fact—only assembled after a month of anxiously marking time rather than before it began. Hence a practical conclusion worth taking away from this chapter on its own: a closed list of risks is best drawn up on the first day of work on a major decision, not the last.

Step three. Set a fixed deadline for the Search phase and hold to it as a commitment, not a wish. The team was given twelve days to close all six items on the checklist. Twelve days is not an arbitrary number: that was exactly how much time remained until the decision had to be passed to the bank, with a margin for legal paperwork before exclusivity expired. It’s worth highlighting the management principle behind this step: the deadline for the Search phase is set not by a feeling of “when I’ll be ready” but by an external constraint—a market window, the board’s decision cycle, the date after which the cost of inaction starts growing faster than the cost of a possible mistake. Had Igor Valentinovich kept waiting for an inner sense of readiness, it might never have come—because the feeling of “checked enough” does not appear on its own where a painful experience once happened; it arises only when the checking process itself stops.

Step four. Separate the decision from responsibility for the past mistake—explicitly, out loud. At a separate short meeting with the chairman of the board, Igor Valentinovich named outright what had gone unspoken until then: the deal of three years before was still shaping his attitude to any major acquisition. The chairman responded concretely rather than with general words of support: he listed three differences between the old deal and the current one. There, the contracts had not been checked directly with the end customers; here, they had. There, the independent technical assessment of the equipment had been a formality; here, it was detailed, with a site visit. There, the decision had been made in five days under pressure from the seller; here, it was being made according to a plan with a checklist. The conversation took twenty minutes. Psychologically, it did the same thing that Klaus Scherer’s appraisal theory describes as a sequence of checks of a situation: novelty, relevance, control. As long as the past deal and the current one were perceived as a single category—“major acquisitions”—control seemed low, because last time there hadn’t been enough of it. As soon as the differences were named concretely, the situation stopped belonging to the same category, and the part of the anxiety tied specifically to the resemblance to the past subsided.

Step five. Mark the moment of transition from Search to Decision as a separate, visible event. On the twelfth day, when all six checklist items had been closed, the company held not an ordinary working meeting but a separate session explicitly designated as the decision point—with the board, the CFO and the division head taking part. The format was pointedly different: not “let’s go over the details once more” but “on the basis of the closed checklist, we decide yes or no.” This is not a bureaucratic formality. The difference between the Search phase and the Decision phase in the emotional cycle is the difference between gathering information and moving to a binding action, and people tend to get stuck at the border between them if that border is not clearly marked. Richard Lazarus wrote about exactly this: once the appraisal of a situation is complete and resources are judged sufficient, readiness to act must give way to the action itself—otherwise the appraisal starts going round in circles without ever turning into behavior. A separate meeting with an unambiguous agenda served as exactly that kind of border.

The decision was made at that meeting: “yes,” with one condition—to lower the offered price by 6%, since the technical assessment had found that part of the equipment had a shorter remaining service life than the seller had stated. The offer was sent to the bank on the forty-third of the forty-five days allotted—on time, without asking for an extension.

It’s also worth explaining why this protocol is not an invention from scratch but a special case of a more general management principle that scientists and practitioners have discovered independently in various forms. The psychologist Daniel Kahneman and his co-authors Olivier Sibony and Cass Sunstein showed in their book “Noise” (2021) that a significant share of the inconsistency in executives’ judgments is explained not by bias—a person systematically erring in one direction—but by “noise”: variability in judgments that should agree—different people, or the same person at different times, assessing one and the same situation differently. Their practical conclusion is “decision hygiene”: structured procedures set in advance reduce this variability better than an attempt to be “more objective” through sheer willpower. A checklist with a closed list of items and a fixed deadline is the simplest form of such hygiene: it does not remove anxiety as an emotion, but it prevents anxiety from deciding anew each time when exactly the checking process can be considered finished.

The result

All the figures below are illustrative—a composite example for a diversified industrial holding company of comparable scale, not the performance of a specific existing company.

The deal closed forty days after the bank accepted the offer, allowing for the time needed for legal paperwork. Already in its first full year within the group, the acquired asset took second place by revenue in the energy services division, adding about 34% to the division’s revenue. Both key contracts with industrial customers—the ones the team had specifically traveled to confirm in person at the checklist stage—were renewed for a new term under Northern Perimeter’s management. In other words, the fear that a change of ownership would scare customers away did not materialize, because it had been checked in advance rather than after the fact.

More telling is what happened to the decision-making process itself in the holding company. The six-item checklist and the rule of a fixed deadline for the Search phase, devised for one specific deal, were formally adopted by the board as a mandatory procedure for any investment decision above a certain threshold. In the year and a half after this deal, the holding company completed three more deals of comparable size. The average time from the appearance of a proposal to the board’s final decision fell from the previous open-ended timelines (in practice, from 45 to 90 days depending on the deal) to a consistent 12–16 days, with the same scope of checks—just without the repeats. None of the three deals was subsequently judged a mistake in the annual review of the investment portfolio.

These illustrative figures are consistent with the overall picture that real market statistics show. The McKinsey study mentioned at the beginning of the chapter puts the total time managers lose to ineffective decision-making at roughly 530,000 person-days a year for a typical large company on the scale of the Fortune 500 list—not because of wrong decisions but precisely because of the time teams spend circling around a decision that sooner or later will be made in the same form anyway. In this logic, what Northern Perimeter saved was not the showiness of one deal but something more valuable: predictable, reproducible time for the whole organization at future crossroads of similar scale.

It is also worth noting that the fixed-deadline checklist that Northern Perimeter’s board reinvented for its own needs essentially reproduces a principle that strategy consultants formalized earlier under names of their own—for example, the allocation of roles in decision-making that the consulting firm Bain describes with its RAPID model (Recommend, Agree, Perform, Input, Decide). The idea in both cases is the same: a decision stops stalling not when its complexity decreases but when it becomes unambiguously clear who exactly is obliged to say “yes” or “no,” and when. Before the protocol was introduced, this role was personal rather than structural—Igor Valentinovich was at once the one assessing the risk, the one who had to make the decision, and the one bearing sole responsibility for the memory of the past mistake. Combining all three roles in one person without an external structure is in itself a factor that lengthens the Search phase: there is no one to share the burden of assessment with, no one to lean on if the anxiety proves excessive. Shifting at least part of that burden onto a separate checklist, fixed on paper with external deadlines, took some of the load off the personal, emotional side of the decision—and handed it over to the procedure.

It is also worth recording what did not change—this matters, so as not to create the impression that the protocol rids you of anxiety altogether. Igor Valentinovich continued to feel pronounced tension before every major deal even after the protocol was introduced. He himself later described it to the strategic development team this way: “It’s exactly as frightening as before. It’s just that now the fear gets twelve days, not thirty, and not forever.” This is a precise description of what the protocol gives: not the cancellation of an emotion but a boundary for its action. Fear remains the same extremely useful measuring instrument it was from the very beginning—what changes is only how long the organization allows it to take the wheel instead of simply showing the readings on the dashboard.

Burnout in a company’s top executive is almost never a personal weakness or a matter of character. It is a management gap: the company has not built a rhythm that forcibly brings the leader back from a phase of nonstop action into a phase of recovery. As long as that rhythm is the director’s private affair, dependent on willpower and mood, it loses sooner or later: willpower is used up faster than it can recover, and precisely when it is needed most. The task facing the head of a second-generation family business is not to “learn to rest” or to “find balance,” as popular self-help books advise. The task is to build recovery into the very architecture of how the company is managed: into the calendar, the allocation of authority, the reporting system—just as financial accounting, occupational safety or the warehouse inventory schedule are built in. Then the top executive’s capacity to work no longer depends on remembering, yet again, the word “enough,” and becomes what it should have been from the start—a managed business parameter.

The situation: a company where no one gets to rest

A chain of twelve consumer electronics and home appliance stores in a regional capital and three neighboring towns. It was founded in the mid-nineties by the current director’s father, who started with a single stall at a market, had built three stores by the two-thousands and by the twenty-tens had handed management over to his son, keeping for himself the formal status of honorary chairman, which, admittedly, he hardly ever used. The son—in this composite, illustrative story, let’s call him Andrei Dmitrievich—took over the company at thirty-two and in ten years grew the chain to twelve locations, moved into online sales, built his own logistics and warehouse, and came through two economic downturns and one change of main home appliance supplier after some foreign brands left the Russian market. The company is consistently profitable, with a recognizable name in the region and loyal customers who still remember his father’s store at the market and tell salespeople about it as a family legend.

By the winter of twenty twenty-five, going into twenty twenty-six, Andrei Dmitrievich is forty-two. He has not taken a vacation in more than three years—unless you count two short trips with his family, during which, by his own admission, he was “formally at the beach, but actually on the phone.” He can’t remember the last time he switched his phone off completely for a weekend; most likely it was while his father, who died four years ago, was still alive. His working day starts at seven in the morning with messages from store managers—usually questions the store managers could perfectly well resolve themselves but prefer to play it safe and ask about—and ends with a call with suppliers, for whom, in their time zone, the evening is only just beginning. Saturdays mean touring the stores, because “seeing it with your own eyes is more reliable than reading a report.” Sundays mean going through the weekly financial report, because on Monday morning he has to face the store managers with ready decisions, not with questions—or at least that is how Andrei Dmitrievich explains the point of this routine to himself.

From the outside, all this looks like an ordinary, even familiar story of a successful workaholic—the business press has plenty of such stories, and they are often held up as a model of managerial dedication. The problem lies elsewhere, and it shows clearly in the numbers, not just in how the director feels. Over the past year and a half, three people whom Andrei Dmitrievich had personally spent years grooming as his deputies left the company one after another—two of them went to direct competitors, taking with them an understanding of how the chain’s purchasing and pricing work. Revenue grew by only four percent—while the regional home appliance market grew by nine over the same period, according to the industry association’s estimate. The director himself was laid up twice that winter with a hypertensive crisis, which everyone in the company knows about—in a structure like this, rumors spread faster than any official announcement—but it is not done to discuss it openly: neither with Andrei Dmitrievich himself nor among colleagues at operational meetings. At the last annual meeting with the managers of the twelve stores, he fell silent in the middle of the discussion for a good half minute—in a packed room of twenty-odd people—unable to put into words a simple point about the margin plan for the next quarter, even though that figure had been in his head that very morning. For two days afterward he put it down to fatigue and lack of sleep, although everyone in the room understood they had seen something more serious than ordinary tiredness from a heavy schedule.

His mother, who formally stepped back from the business when management was handed over but still holds ten percent of the company and attends the board once a quarter—a family tradition laid down by his father—said one thing to him after that meeting, in a one-on-one conversation, and it became the trigger for examining the situation: “At your age, your father had already forgotten the word ‘must’—he knew how to tell himself that the work could wait until Monday. You, it seems, have forgotten the word ‘enough.’” The remark came across not as a reproach but rather as a diagnosis from someone who had been watching the company for decades and saw a pattern where Andrei Dmitrievich himself saw only the day-to-day workload.

Diagnosis through the cycle: where exactly the movement got stuck

To understand what actually broke here, it helps to recall how the cycle of emotional regulation works—the cycle discussed in detail in Part I of the book. A person has four phases that normally follow one another, closing into a loop and starting again at a new level of the task. Search—when the brain assesses the situation and looks for a goal significant enough to be worth spending energy on. Decision—a short phase of choice, often colored by anxiety, because anxiety, according to Joseph LeDoux’s neuroscience of fear, is the work of the amygdala, a small structure deep in the brain that instantly, even before conscious thought gets involved, signals a threat and makes us weigh the risk before acting. Action—the phase of productive tension and movement toward the goal, physiologically linked not with avoidance but, as the psychologist Eddie Harmon-Jones showed in his experimental work, with approach: anger and mobilization push a person forward, toward the goal, not back, away from it. And finally, Reward—the phase in which the body registers the result and gets a release of dopamine, the nervous system chemical responsible for the feeling of satisfaction from an achievement. But, and this is what matters for this particular story, the dopamine system, as research by the neurophysiologist Wolfram Schultz and his school shows, responds not to the reward as such but to the difference between expectation and outcome—the so-called “reward prediction error.” It is because of this mechanism that the joy of achievement fades quickly and a person returns to a searching state, but now at a new, slightly higher level of the task. This is normal and even useful—that is how any healthy cycle of development works. The pathology begins not where the cycle speeds up but where one of its phases stops completing altogether.

The key moment of the cycle that is missing from Andrei Dmitrievich’s story is not Search, not Decision, and not even the Action phase itself. What is missing is the boundary between Action and Reward. He physically has no moment of completion. Every closed task—opening a twelfth store, a quarter successfully navigated, a conflict with a major supplier resolved after a failed delivery—does not become a point where the brain registers “done, I can breathe out”; instead, it instantly, without a pause, turns into the starting point of the next task. In her “broaden-and-build” theory of positive emotions, Barbara Fredrickson shows that positive emotions—above all the joy of achievement—briefly broaden a person’s repertoire of possible actions, give them more creative freedom for a short time and help consolidate the experience gained as something to build on later. But in Andrei Dmitrievich’s case this phase physically has no room left: as soon as a store opens, a report on a stock shortage at the warehouse of a neighboring location arrives that same evening, and the brain jumps straight into a new Decision phase, skipping the registration of the previous cycle’s result entirely.

This is the technical, not metaphorical, definition of chronic burnout seen through the lens of the cycle: not “too much work” as such—people endure intense work for years without burning out, provided the cycle is completed in full—but a break in the cycle at one specific point, between Action and Reward, repeated hundreds of times in a row without a single exception. In their theory of the self-regulation of behavior, Charles Carver and Michael Scheier describe how human behavior is regulated through a constant comparison of the current state with a goal, and when a goal either becomes unattainable or its achievement is never confirmed to the body as an accomplished fact, the person does not go through the normal stage of “goal disengagement”—a state close to calm and a meaningful pause, which normally precedes setting a new goal, one consciously chosen rather than picked up by inertia simply because the previous task has not yet been formally closed. Without this pause, each next goal is taken on not by a fresh body ready for it but by one already exhausted—and the threshold beyond which decisions get worse, memory fails in front of other people and the body breaks down in hypertensive crises arrives predictably, and sooner or later it arrives for everyone who lives in this mode for years.

There is also a second level of breakdown in this story—organizational, not just personal. The QAC model (Quantum Ambidexterity Cube, the author’s model of an organization’s management maturity, discussed in detail in the second chapter) describes the Emotions domain as one of the measurable management domains alongside Power, Tech&Data, Processes and Culture—that is, as something that can and should be managed systematically, not as a “soft” topic left to each employee’s personal responsibility. In Andrei Dmitrievich’s company this domain sits at the lowest, “ad hoc” maturity level: the top executive’s recovery rhythm is not a subject of management at all—not for the board, not for the director himself, not for the HR department, which keeps track of vacations for every employee in the chain except one. Everyone sees it as a matter of his personal discipline and character, not as an operational risk for a second-generation company, although in essence it is exactly the same kind of risk as the failure of the single server on which the chain’s entire accounting runs. If the director is the only point through which the final decision on most issues in twelve stores physically passes, then the exhaustion of that one point is the exhaustion of the entire management system—and an exhaustion that builds up unnoticed and surfaces abruptly, as it did at the annual meeting.

It is worth taking a closer look at why three potential deputies left the company in a year and a half—a figure alarming enough in itself for a chain of this size not to be written off as chance or as general turbulence in the labor market. A closer look—exit interviews conducted by an outside consultant and separate conversations with the managers who stayed—revealed that all three, at some point in their time at the company, had received a substantial chunk of authority from Andrei Dmitrievich—the right to negotiate independently with a major supplier, the right to approve retail discounts, the right to set staff schedules in their area—and then that authority was quietly taken back, without any announcement. Not because the deputies made mistakes—on paper they had few errors to their name—but because the director, stuck in a phase of unending Action, physically could not stop and trust someone else’s decision without immediately double-checking it after the fact. Team psychological safety, which Harvard Business School researcher Amy Edmondson describes as essential in her work on effective teams, includes a subordinate’s confidence that a decision delegated to them will remain their decision and will not be quietly revised from above at the first opportunity. When the top executive does not go through the Reward phase and does not register for themselves the fact that “this task is closed, and closed not by me personally but by my deputy, and that is enough,” they unconsciously devalue other people’s results—simply because they do not know how to devalue, that is, truly let go of, their own. One of the deputies who left described it to the consultant almost word for word: “I didn’t leave because of the money; we had a decent salary. I left because in six months I was never once allowed to see an issue through to the end myself—Andrei Dmitrievich always came back and checked, as if I were an intern and not someone he had chosen himself.”

A similar picture emerged from the statistics of the meetings themselves—the same kind of meeting at which the director had fallen silent for half a minute. Going through the minutes of operational meetings for the past six months, the outside consultant noticed that almost ninety percent of the issues at these meetings ended with Andrei Dmitrievich’s phrase “fine, I’ll think about it and decide myself”—even when the manager of a particular store had already proposed a ready, sensible solution and was only waiting for formal approval. Formally, it looked like the top executive taking responsibility. In substance, it was a symptom of the same break in the cycle: the inability to accept someone else’s decision as final, because his own Reward phase does not work and does not send the signal “this part of the system functions even without my personal involvement.”

The solution: institutionalizing the recovery rhythm

The first thing that became clear at the diagnostic stage: individual advice like “take a vacation,” “delegate more” or “find time for yourself” would not work here—he had heard it more than once from his wife, from his mother and from a doctor he knew whom Andrei Dmitrievich had seen after his first hypertensive crisis, and not once had he followed it for longer than two weeks in a row. The problem was not a lack of knowledge that he needed to rest—Andrei Dmitrievich knew that perfectly well and could recite the right words about work-life balance himself. The problem was the lack of a structure that makes rest not a one-off act of will by a tired person—and therefore the person least able to summon that will—but part of a system that works regardless of his state and mood at the moment, on any given Tuesday. The solution was built in five steps—from an honest measurement of the current picture to a fixed, written policy that would outlive the very thought that it could be abandoned at any moment.

Step one. Measurement instead of feeling. Before changing anything in how the company was run, the vague “I’m tired” and “this can’t go on” had to be turned into concrete numbers that could be discussed, because you can truly manage only what is measured, not what is felt. For four weeks, at the outside consultant’s suggestion, Andrei Dmitrievich tracked three simple parameters—not in a special app but in an ordinary table he filled in in the evening before bed. The first parameter was the number of hours a day when he was physically unavailable for work matters: not “asleep,” but actually unavailable, meaning his phone was switched off or lying in another room, weekends included. The second was the number of decisions a day that he made personally rather than delegating them to one of the store managers or deputies; not only the number was recorded but also each decision’s level of significance—from “what discount to give a regular customer” to “should we change our home appliance supplier.” The third was a subjective rating of his own state on a simple five-point scale, twice a day, morning and evening, always noting not only the intensity of the fatigue but also its sign: irritation versus apathy—these, as the consultant explained, are fundamentally different states that call for different management measures, not the same advice to “get some rest.”

Over the four weeks, the picture confirmed the initial diagnosis quantitatively, not just in how things felt. Unavailability averaged forty minutes a day, a little over half of which was time in the shower in the morning—the only place his phone physically did not go. More than eighty percent of operational decisions across all twelve stores—from pricing questions to staff reshuffles at individual locations—passed through Andrei Dmitrievich personally one way or another, even though there formally were store managers whose job descriptions expressly gave them that very right. His evening rating almost never rose above “tolerable,” and on average twice a week it dropped to a level that Andrei Dmitrievich himself described in a comment in the table as “on autopilot”—a state in which decisions get made, but without real attention to their consequences.

Step two. Shifting responsibility from the person to the calendar. Instead of a promise to himself to “try to work less”—a promise that, as the earlier experience with his wife’s and mother’s advice had shown, shattered against the first serious work issue—fixed blocks of the top executive’s unavailability were entered into the company calendar. And not into Andrei Dmitrievich’s personal calendar, but into the company’s shared calendar, visible to all twelve store managers and to the board: two full weekday evenings a week, all of Saturday, and once a quarter, seven consecutive calendar days with no access to work communication channels at all, including the messenger app where all of the chain’s operational fires usually played out. The difference here is fundamental, and it is precisely where the mechanism of institutionalization lies—the mechanism this chapter has been talking about since its first paragraph: this is not Andrei Dmitrievich’s personal plan, which can be quietly shifted to deal with the latest fire at one of the stores, but an organizational decision recorded in the minutes of the board of directors—as binding as the quarterly warehouse inventory schedule, which no one would postpone simply because it is inconvenient for the warehouse clerk that day. Breaking one of these blocks stopped being the director’s own silent decision of “oh well, never mind, I’ll answer”—it now required a formal justification to the board: why this particular emergency justifies breaking the policy, and what has been done so that a similar situation does not happen again.

Step three. A distributed decision point instead of a single one. For the unavailability blocks to become physically possible at all, rather than remaining a nice entry in the calendar that would be broken three times a week anyway, the root cause of the overload had to be addressed: eighty-odd percent of decisions converged on one person. The twelve stores were grouped into three territorial clusters of four locations each, and each cluster was given a manager—some drawn from existing store managers who had proven themselves over the years, some hired from outside—with a clearly written right to make decisions without the director’s sign-off: on prices within a set percentage band, on staff schedules at their four locations, on small purchases of supplies and equipment for the stores, on settling disputes with customers up to a certain amount of compensation. The key difference from the earlier delegation attempts, already tried and failed—the very ones that had cost the company three deputies—was that the new authority was set down in a written policy with specific numerical limits rather than a verbal arrangement along the lines of “sort it out yourself, I’ll give you a hint if you need anything.” The written boundary was needed not so much by the three new managers as by Andrei Dmitrievich himself—it physically stopped him from coming back to “offer a hint, just in case,” because the issue was now formally outside his remit rather than simply outside his current attention, which, as the experience of the three departed deputies had shown, had a habit of returning at the worst possible moment.

Step four. A completion ritual instead of an endless stream of tasks. Separately, following a conversation with the consultant, a simple weekly ritual was introduced: every Friday at the end of the working day, twenty minutes that Andrei Dmitrievich spent alone, with no meeting and no phone at hand, answering three questions in writing in the same notebook where he had earlier kept the measurement table. The first question: what was seen through to the end this week and can be considered closed, with no caveats at all. The second: which of the things done deserved real attention and effort, and which were simply the background noise of the day that took up time but not energy. The third, by Andrei Dmitrievich’s own admission the hardest: what can be consciously let go right now—not because the task is finished, but because on reflection it is not a priority and will not become one in the foreseeable future. This is a direct, essentially engineering solution to the problem this chapter has examined above—the missing Reward phase in the cycle. The ritual does not replace proper rest and does not claim to, but it creates the very boundary Carver and Scheier wrote about—a concrete, repeatable point where the brain can formally register “this task is closed” before moving on to the next week, instead of rushing on by inertia without a single visible marker along the whole way.

Step five. The same rhythm for the whole management team, not just the top executive. This is where the decisive shift happened—from one tired person’s personal practice to the management practice of an entire company: by a board resolution, the rule of fixed unavailability and the weekly completion ritual was extended to all three new cluster managers rather than left as a special condition for the director alone. This solved two problems at once, both of them managerial in substance. First, it removed the very real risk that within a year or a year and a half it would not be Andrei Dmitrievich who burned out—he was now under the watchful eye of the policy—but his new deputies, who had taken on eighty percent of the director’s former decisions and risked repeating his path exactly, just one level down. Second—and this, as the following year showed, turned out to be even more important from a purely managerial point of view—extending the policy to the whole team turned the recovery rhythm into a company norm rather than a special, almost privileged condition for the top executive, which other employees might read as “the bosses can, but we can’t.” When a cluster manager sees with their own eyes that the director really is unavailable on Saturdays, and that this does not cause a crisis, does not tank sales and does not lead to a panicked call at seven on a Sunday morning, they get permission of their own—not a declarative one, read in a corporate policy, but one lived through their own observation—to hold their own boundary with the store managers who report to them.

The question of the board of directors—or, more precisely, of Andrei Dmitrievich’s mother as the holder of a ten-percent stake and the living voice of his father’s generation in the company—called for a separate, delicate solution. A second-generation family business carries a specific burden here that is not always obvious from the outside: blood relatives easily read the top executive’s rest—not out of malice, simply because of family history—as a lack of devotion to the business that the previous owner literally built from scratch. The conversation with his mother was made a separate, preplanned item on the board’s agenda rather than left to a chance family chat over dinner. She was shown those same four weeks of measurements—forty minutes of unavailability a day, eighty percent of decisions resting on one person—as well as the management turnover figure for the year and a half and the data on the company lagging behind industry growth, and she was invited to see the unavailability policy not as an indulgence of a weak character but as a condition for keeping the company in working order over the next ten to fifteen years—that is, as directly in her own interest as a shareholder and as someone for whom the fate of the company his father built remains a personal matter. Formal approval by the board mattered here not so much legally—in a structure like this, the board is more of an advisory body than one with real enforcement powers—as psychologically: it relieved Andrei Dmitrievich, once and for all, of the need to fight for his right to a pause personally and anew each time in front of the family, which would have turned a one-off difficult conversation into a constant source of guilt.

The result: what changed over the year

What follows are illustrative figures. This is a composite, fictional example based on the typical picture for companies of this size and profile, not the reporting of a specific existing chain—the numbers here show the order of magnitude and the direction of the impact, plausible for the industry, not a precise measurement of one real company.

Twelve months after the policy was introduced, the share of decisions converging on the director personally had fallen from over eighty to about thirty percent—the bulk of operational issues was now being closed at the level of the three cluster managers, and without a noticeable rise in errors or customer complaints, which was what Andrei Dmitrievich himself had feared most. The number of hours the director was physically unavailable for work matters had grown from forty minutes a day to almost two and a half hours on weekdays, plus full days off on Saturdays—three or four Saturdays out of four a month, not counting the separate quarterly week-long break, which was now spent genuinely without a single work message. Contrary to Andrei Dmitrievich’s long-standing fear, not one of these breaks led to any noticeable operational disruption in the chain—nothing like the disaster whose prospect had for years kept him from even trying to truly disconnect from work channels.

Management turnover—the very metric with which this whole conversation effectively began after the annual meeting—changed more noticeably than the other indicators. Over the year, not one of the three new cluster managers left the company, whereas in the previous year and a half three people of comparable standing as deputy directors had left. The chain’s revenue grew by eleven percent over the year—against four percent a year earlier, while the industry in the region grew by around eight to nine percent; in other words, for the first time in two years the company outpaced the market rather than steadily lagging behind it, as had been the case all the while the director physically had no time for anything but the latest fire. One more indicator is worth recording, one that is not financial in itself but is directly tied to the risks this story began with: over the year there was not a single hospitalization or acute episode like the hypertensive crises of the previous winter, and the evening self-rating—the same simple five-point scale, which Andrei Dmitrievich kept up on his own initiative, without reminders from the consultant—held steadily in the range between “okay” and “good” instead of the old constant swings between “tolerable” and “on autopilot.”

One detail is telling as well—it does not show up directly in the financial statements, but it says a lot about what really happened. About six months after the policy took effect, one of the three cluster managers independently, without any instruction from above and without even discussing it with the director, introduced a similar rule of fixed weekend unavailability for the four store managers in the cluster—simply copying the model they had seen working one level up. The institutionalization of the recovery rhythm worked here not only vertically, top-down, through a formal board resolution, but also horizontally—as a cultural pattern that began to reproduce itself without any directive effort from above, which, perhaps better than any figure in a report, shows that the practice took root not as a one-off measure but as a new norm of work.

A similar picture, though not identical in its details, appears in Deloitte and Workplace Intelligence’s research on burnout among senior executives (2022–2023, more than three thousand one hundred participants): around seventy to seventy-five percent of executives had seriously considered leaving their position for the sake of their own well-being—in other words, Andrei Dmitrievich’s problem was not a peculiarity of one family company but a particular case of a widespread phenomenon. The recovery experience model of the psychologists Sonnentag and Fritz points the same way: of the four channels of recovery—relaxation, mastery (learning something new), a sense of control over one’s time, and psychological detachment from work—it is complete psychological detachment that turns out to have one of the strongest effects, stronger than rest during which one’s thoughts stay at work. This directly explains why Andrei Dmitrievich’s earlier attempts to “rest,” formally at the beach with a phone in his hand, produced no result: being physically on vacation without psychologically detaching from work simply is not a recovery experience. The Russian picture is telling in this respect too: according to a Kontakt InterSearch survey (2024, five hundred forty-seven executives), eighty-nine percent of Russian top managers consider burnout not a personal problem but a full-fledged business problem, and seventy-five percent have experienced it personally—that is, the very framing of the question with which this chapter opens is shared by the overwhelming majority of executives, even if far from all of them arrive at a systemic solution like the one built in this story.

When a bank announces that it is rolling out artificial intelligence tools (AI—systems capable of performing tasks that used to require a human: analyzing documents, answering customers, preparing loan decisions), the first thing to look at is neither the architecture of the solution nor the project budget. Look at the middle management layer—the branch heads, heads of business lines and senior managers who stand between the board of directors and rank-and-file employees. They are the first to go through the fear phase that the whole workforce faces, and whether the technology rollout turns into growth or a rift depends on whether senior management sees this phase in time and accepts it as legitimate rather than as “resistance to change.” This can be managed—but not with motivational speeches or by pulling deadlines forward. It takes a systematic map of the organization’s emotional climate, work on psychological safety and a leader’s ability to recognize what is going on: right now the whole team is in the fear phase, and it has to be handled as a phase, not as sabotage.

The situation: a quiet rebellion instead of open conflict

A mid-sized retail bank—several million customers, an extensive branch network, the industry-standard range of products: deposits, loans, cards, mortgages—decides to bring AI tools into its lending pipeline and its customer service. The management board’s logic is clear and, in its own way, flawless: competitors are already automating application scoring (assessing a customer’s creditworthiness), chatbots are taking over routine inquiries, and the labor market is such that every hour of an employee’s time freed from routine is a direct saving. The board of directors approves the program, allocates a budget and hires a systems integrator. From the standpoint of the technology and the financial model, everything has been done right.

Three months after the pilot starts, the project gets stuck. Not because the algorithm doesn’t work—the algorithm is fine, and scoring accuracy is above plan. It gets stuck because middle management—the heads of lending departments, the heads of problem-loan teams, the senior tellers—is sabotaging the rollout in the way any leader finds most awkward: not by refusing openly but by letting it slowly fade. Meetings are polite. Implementation reports are submitted on time. But the number of applications that “for some reason” are processed manually, bypassing the new tool, keeps growing. The proactive suggestions that line managers used to put forward by the dozen every month dry up completely. In meetings with senior management there is compliant silence; in the smoking areas and the department’s corporate chat the tone is entirely different, and it happens to catch the HR director’s eye through a complaint from one of the few employees who have spoken up openly.

The management board reacts the way almost any management board reacts in a situation like this: it tightens control. It introduces mandatory targets for the share of applications processed through the new tool. It appoints people responsible for a “culture of embracing change”—a phrase that in itself signals that the conversation is being conducted in the language of demands, not of understanding. The result is the opposite: the manual bypass rate does not fall but rises, and staff turnover starts rising after it—precisely at the level of the “player-coaches,” the very heads of departments and teams without whom the pipeline physically cannot work, because they are the ones who deal with exceptions, non-standard cases and live customers rather than rows in a database.

One telling detail usually escapes a management board focused on aggregate metrics. At the level of an individual branch, sabotage almost never looks like a coordinated action. A department head who sends an application for manual processing “just in case” sincerely does not see themselves as part of any resistance—they see themselves as a responsible manager protecting the customer from a mistake by a new system that has not yet stood the test of time. The problem is not that this explanation is disingenuous but that it is only partly true: nine times out of ten, behind the reasonable caution lies a much simpler motive that this manager will most likely not put into words even to themselves—the anxiety that the better the algorithm works, the less point there is to their own position in the company. Separate, scattered acts of “reasonable caution,” multiplied across several hundred branches, are what create something that, at the level of a report to the board of directors, looks like the organization’s mysterious inability to absorb an obviously useful technology.

One more detail deserves separate attention: the climate deteriorated past the point of no return faster than the management board could find out about it. The bank’s standard reporting system was set up so that warning signals from the branch level reached the board of directors through two or three layers of management—and at each layer the message was smoothed over, because no intermediate manager wanted to report a problem in their own area upward without a ready solution. By the time the manual bypass figures finally made it into the consolidated project report, real anxiety in the branches had been building for two months. This is not a coincidence either but a systemic consequence of the lack of psychological safety: it stifles truthful information not only between a rank-and-file employee and their manager but at every subsequent rung of the management ladder, including the one where the HR director sits (the executive responsible for people and corporate culture across the entire organization).

This is exactly the point at which it is worth stopping and asking the question differently from the way it is usually asked: not “why are they resisting the rollout?” but “in which phase of the emotional cycle has this part of the organization got stuck, and what exactly is holding it there?”

Diagnosis through the cycle: what exactly got stuck

Let us briefly recall the logic examined in detail in Part I of the book. A person—and, adjusted for scale, an organization as well—moves through an emotional cycle of several phases: Search (finding one’s bearings in a new task, often accompanied by anxiety), Decision (the move to certainty, choosing a direction of action), Action (productive tension, working toward the goal) and Reward (closing the cycle, recognition of the result, after which comes Search again, but at a new level of task complexity). This sequence has a physiological basis: the amygdala—a brain structure that, according to the neuroscientist Joseph LeDoux, assesses potential threat quickly and before conscious reasoning kicks in—triggers a state of anxiety before a person has consciously understood what exactly the new technology threatens. The psychologist Jeffrey Gray showed that moderate anxiety sharpens attention and mobilizes, but once a threshold is crossed, the behavioral inhibition system switches on—and then a person does not just hesitate; they physically stop, freeze and avoid action. The difference between “wary attention” and “paralysis” lies not in a person’s character but in the intensity of the threat signal and in whether the person feels they can influence it.

In our bank, middle management entered the Search phase—and got stuck precisely in its anxious rather than its productive part—for three overlapping reasons, none of which would have been fatal on its own.

The first reason is existential uncertainty that was never spoken aloud. Not a single official project document answered a simple question: what will happen to the role of the head of a lending department in two years if the algorithm does most of what that person does today? Management’s silence on this question is not a neutral position. The psychologist Richard Lazarus, author of the cognitive appraisal theory of emotion, showed that an emotion arises not from the event itself but from how a person appraises its significance for their own goals and whether they have the resources to cope. In the absence of information, people do not stop appraising the situation—they appraise it for the worse, because uncertainty in itself is interpreted as a threat. The gap in the management board’s communication was filled with rumors of upcoming layoffs—not because anyone on the board was actually planning layoffs (at that stage, staffing decisions were not even being discussed) but because a vacuum of meaning is always filled with an anxious scenario if nothing else fills it.

The second reason is the lack of psychological safety—that is, as defined by Harvard Business School researcher Amy Edmondson, team members’ confidence that they can speak openly about mistakes, doubts and uncomfortable observations without risk of punishment or damage to their reputation. In this bank, that confidence had been missing even before the AI tools arrived: the culture was built on projecting confidence and wrapping up meetings quickly without any public doubts—“if you have a problem, you’re not coping.” In calm times this feature of the culture stayed in the background and was not critical. At the moment when the whole organization needed to admit, “we don’t know how this will work, and that’s normal,” it became fatal. Department heads could not say “I’m afraid” openly, because in this culture fear was equated with weakness and professional unfitness. Google’s large study known as Project Aristotle (2015) showed, on data from more than 180 work teams, that the main factor in a team’s effectiveness is not its mix of competencies or individual talent but precisely the level of psychological safety within it. A team in which you cannot admit to anxiety does not become braver—it simply stops telling the truth upward and goes on being afraid in silence.

The third reason is that the very design of the rollout program ignored middle management’s function as the cycle’s shock absorber between the board of directors and frontline staff. The management board had gone through its own Search and Decision phases back when the budget was approved—for the directors, the decision had been made, and what came next was the Action phase: implement, report, speed up. But for the department heads and their staff, the Search phase was only beginning at the very moment when the management board was already demanding Action-phase metrics from them. This is the classic managerial paradox of change that the researcher John Kotter described: organizational transformations fail not so much because of strategic errors as because leadership does not take into account the emotional path people must travel before a strategy becomes their own decision rather than an order handed down from above. The management board was demanding productive tension from people who were physiologically still in a state of anxious orientation. Demanding action from a person in the Search phase is like demanding that a runner speed up while they still haven’t figured out which way to run.

Put these three reasons together, and what you get is not “resistance to change” in the vague HR sense of the phrase but a precise diagnosis: a collective fear phase, amplified by an information vacuum, with no channel for legitimate expression because of the lack of psychological safety, and pushed along by leadership demanding the results of a phase the team had not yet reached. That is exactly why tightening control—mandatory targets, disciplinary conversations—did not work and could not have worked: it hit the symptom (manually bypassing the system) but did not remove the cause (an unfinished, locked-in fear phase), and by its physiological nature it only raised the intensity of the threat and thereby reinforced the avoidance response.

It is worth pausing here on why the intuitively most obvious management response—“show people with numbers that there is nothing to fear”—proved useless, even though the management board’s arguments were essentially correct: the algorithm really did reduce the number of errors in assessing applications and really did free up time for more meaningful work with customers. The point is that the fear phase, by its physiological nature, as the neuroscientist Joseph LeDoux showed back in the mid-nineties, is triggered by brain structures that work faster and more crudely than the regions responsible for rationally weighing arguments. The amygdala responds to a threat signal before the prefrontal cortex (the part of the brain responsible for deliberate reasoning) has had time to process the argument “the algorithm is more accurate than a human.” Persuasive statistics addressed to reason physically cannot directly switch off a response triggered by an older and faster system of the brain. You can get through to it not with a more persuasive argument but by reducing the threat signal itself—that is, by working on the source of the anxiety rather than on the logic of resisting it. That is why dozens of presentations with efficiency charts did not move the situation by even a percentage point until the very nature of the signal the department heads were receiving had changed.

The solution: an emotional climate map instead of an order

The turning point came not when the management board decided to “explain things better”—general calls for openness do not work in a culture with low psychological safety, because people need not words about openness but proof that it is safe to tell the truth. The turning point came when the bank built a sequence of five steps, each of which addressed one of the three identified causes of getting stuck rather than the problem as a whole in one go.

Step one. Diagnosing the climate instead of making assumptions about it. Before changing anything in the rollout program, the bank put together what might be called an emotional climate map: a short, anonymous, regular staff mood survey—not a general once-a-year “engagement index” but a frequent pulse survey of a few questions that measured not opinions about the project in general but specific, measurable parameters of the teams’ emotional state: the level of anxiety about one’s own role, the sense of control over what is happening, and trust that management is telling the truth about its plans. The idea that an emotion can be broken down into several measurable components—intensity, valence (whether it is positive or negative) and the speed at which it builds or subsides—and worked with systematically rests on a general principle of affective science (the science of emotion) and organizational psychology, not on any single specific method. The value of this map lay not in the numbers themselves but in what it revealed: anxiety was not spread evenly across the bank but concentrated precisely at the level of the heads of departments and teams—that is, exactly where the algorithm physically meets the live customer, with a non-standard situation that doesn’t fit the model.

A technical detail that no one had thought of at the start also mattered: the bank decided to display the climate map as a simple dashboard (a single screen of key figures) not only for the HR director but for the board of directors itself—once every two weeks, alongside the project’s regular financial indicators. As long as the team’s anxiety was invisible at the same level of management where the budget and deadlines were discussed, it lost out in priority to any financial indicator by definition—not because the directors considered people unimportant but because management attention goes where there is a number and a chart. As soon as the department heads’ anxiety level became a line in the summary sent every two weeks just like the percentage of the rollout plan completed, it had a chance to influence decisions before the situation became irreversible.

Step two. Acknowledging the phase out loud—from the management board’s top executive. The chairman of the management board held a series of meetings, not to present the rollout roadmap but to deliver one specific message: “We understand that for many of you this project is bound up with fear for your future role. That is a normal reaction to a serious change, and we will not pretend it doesn’t exist.” This was not a psychological trick for its own sake—it was a direct application of the idea of psychological safety: until fear is legitimately named from the top, it cannot be voiced from below without risk. At the same time, the management board did what it had failed to do at the start: it gave an honest, if not final, answer to the question of what would become of people’s roles—which functions the algorithm really takes over, which remain and grow in importance (handling exceptions, complex cases, live negotiations with customers), and which new ones appear (tuning the algorithm and monitoring the quality of its work). Not everything in this answer was what people wanted to hear—some operational roles really were going to be cut within a year and a half to two years. But certainty, even unpleasant certainty, is easier to bear than uncertainty with a positive spin—as both Lazarus’s appraisal theory and recent stress experiments suggest: a person can cope with bad news if they know exactly what they are coping with.

Step three. A channel for legitimately voicing doubts—with real consequences. The bank introduced regular closed-door meetings between middle management and the head of the rollout program, without senior bosses present—a format in which one could say “this scenario for the algorithm doesn’t work for our customers” without it being heard as “I don’t want to change.” What was key was not the meetings as such (the bank had had no shortage of feedback formats before) but the fact that some of the doubts raised there led to visible changes in the tool itself—the bank postponed automating several lending scenarios in which department heads had pointed to customers’ life situations that the algorithm did not take into account. This was not a concession for its own sake but a working example of what Edmondson calls a condition of psychological safety: a team must see that honesty is not punished and, what is more, affects the outcome. One such precedent works more powerfully than ten declarations of openness.

One specific case from these meetings is telling; it later became an internal story at the bank, retold without any prompting from the management board. One of the heads of a problem-loan team pointed out that the algorithm systematically underrated applications from older customers with no digital credit history—not because they were less creditworthy but because the model had been trained mainly on data from customers who actively used the bank’s digital services. Formally, this was not a complaint about the rollout but a precise technical observation that the integrator’s engineers could have missed, since they had not worked with this category of customers in person—and it required a direct readjustment of the model. The management board could have treated it as yet another piece of evidence of “resistance” hidden behind a scientific-sounding pretext. Instead, it publicly thanked the author of the observation at a general staff meeting and included the corrected scenario in the next system update. This single case did more for middle management’s openness from then on than the entire corporate chat full of calls for feedback had done over the previous six months.

Step four. Resetting the rollout pace to fit the phase, not the plan. Instead of a single mandatory target for the share of applications going through the new tool across all branches, the bank introduced a differentiated pace: branches and teams whose climate diagnostics showed high readiness (low anxiety, a strong sense of control) got a more aggressive transition schedule and the status of pilot sites, with additional recognition and bonuses. Branches with high anxiety got not an unexplained postponement but a structured, slower path: first, the algorithm worked in advisory mode, with the decision remaining with a person, and only later, as anxiety came down—measured by the same climate map—did the tool move to more autonomous operation. This is a direct application of the idea of self-regulation described in the late nineties by the psychologists Charles Carver and Michael Scheier: movement toward a goal is sustainable when a person keeps a sense of progress and control, not when they are pushed toward the goal faster than they can physiologically adjust. A single forced pace for everyone was an attempt to speed up the Action phase where part of the organization had not yet completed the Search phase—and the second attempt would never have succeeded either had the bank not recognized the right of different parts of the organization to be at different points of the cycle at the same time.

Step five. A rhythm of small wins instead of one final event. The rollout program was repackaged from a single “launch day” into a series of short stages spread out over time, each of which ended with explicit, public recognition: not an abstract thank-you in a mass email but a review, at a general staff meeting, of specific cases in which a department head used the new tool to resolve a customer’s difficult situation faster or better than before. This is a direct use of the mechanism that the neuroscience of reward describes through the concept of “reward prediction error”: the brain’s dopamine system responds not to the reward as such but to the difference between expectation and outcome, and without regular confirmation of progress, spread out over time, the cycle does not close, and the anxiety of the Search phase simply drags on, finding no outlet in the Reward phase. One big finale a year after launch does not produce this effect: the brain needs frequent, tangible points of completion, not a single distant one.

The order of these five steps is not accidental: the climate diagnostics gave leadership facts instead of assumptions; acknowledging the fear phase out loud legitimized what had previously been hidden; a feedback channel with real consequences turned the talk of psychological safety from a slogan into a practice; the differentiated pace lifted the demand for action from those who had not yet gone through the Search phase; and the rhythm of small wins gave the cycle the point of closure it physiologically needs. Each step built on the result of the one before—an attempt to start straight away with the fifth, that is, with celebrating successes, while the first three were still unresolved would have looked fake and only deepened mistrust.

The HR director’s role in this restructuring deserves a word of its own, because it was at this level that the question was settled: would the whole construct remain a one-off PR stunt or become a management system? In this project the HR director stopped being an organizer of engagement events and became the owner of one specific metric—the anxiety level of middle management—reporting on it to the board of directors with the same regularity and the same seriousness with which the CFO reports on margin. This shift in the status of the emotional climate indicator from a “soft topic” to a full-fledged management domain, on a par with financial and operational indicators, turned out to be the structural change that outlived the AI rollout project itself and stayed in the company as a permanent practice.

The result

The figures that follow are an illustrative, composite example typical of banks of this scale and profile, not the indicators of any specific real organization.

Nine months after the program was relaunched along these new lines, the share of applications processed bypassing the new tool fell from a peak of 61% to 9%. The average level of anxiety about one’s own role among heads of departments and teams, as measured by the pulse survey, fell from about 7.2 to 3.4 points on a ten-point scale. Staff turnover at the line management level—that same category of “player-coaches” without whom the pipeline physically cannot work—dropped from 34% to 14% on an annual basis. The number of proactive suggestions from line managers for improving the algorithm’s work, which had practically dropped to zero at the height of the conflict, recovered to a level above the original one, because some of the suggestions were now directly related to tuning and monitoring the tool itself—that is, to the new role, and not only to the old operational work. In the end, the processing speed for a standard loan application increased by about 40%—meaning that the bank ultimately got exactly the economic benefit the rollout had been undertaken for, but it got it nine to twelve months later than originally planned and with far fewer losses among its people than if it had kept pushing on the targets without diagnosing the climate.

It is also worth stressing what did not happen along the way, because the absence of an expected problem is a result too. At the start of the relaunch, the board of directors feared that an honest conversation about cutting some operational roles would trigger a new wave of resignations—that people would leave on their own rather than wait for their role to be cut. This did not happen: turnover fell rather than rose, because certainty, even unpleasant certainty, proved less frightening than continued uncertainty. Some department heads whose former operational functions really did shrink over time used the transition period the bank had honestly announced to retrain for a new function—monitoring the quality of the algorithm’s work and handling complex, non-standard customer cases—and stayed with the company in a new role instead of looking for work elsewhere.

One more outcome deserves a note of its own—it is not directly reflected in the project’s financial model, yet it determines the project’s long-term sustainability: the practice itself—a regular emotional climate map and a differentiated rollout pace—remained in the bank as a permanent management tool rather than a one-off crisis response. The next cycle of technological change—and in retail banking it is inevitable—starts not from zero trust but from an accumulated precedent: management has shown once that it sees fear, names it out loud and changes the pace for the sake of people, not just for the sake of the schedule.

There is a managerial temptation to read this story as a story about communication—as if the management board had simply needed to talk to people better and more often. That is only half true, and it misses the main point: talk without changing the pace of the rollout and without real consequences for the doubts raised would have been yet another declaration that no one would believe after the first mismatch between words and deeds. What works is not talk as such but a specific management discipline: measure the collective phase of the cycle as regularly as you measure revenue; acknowledge the fear phase out loud before demanding productive action from people; and synchronize the pace of your demands with the real, not the desired, speed at which different parts of the organization move through the same cycle.

When a company’s best engineers—not its worst—leave one after another, the first diagnosis is almost always wrong. It seems to be about money: the labor market is overheated, competitors are offering more, so salaries have to go up. That is partly true, but not the whole truth, and usually not the main part of it. In stories like this, money is not the reason for leaving but the last pretext a person names out loud, because they don’t want to discuss the real one. The real reason is almost always the same: the company’s rhythm between challenge and reward has broken. A person has either stopped getting tasks that make them grow or stopped getting recognition proportionate to the difficulty of what they did—and more often both at once. You can keep such an employee with a raise—for two or three months. Then they leave anyway, just slightly better off, and for years the company can’t find an equivalent replacement. To retain people systematically rather than as a one-off, what’s needed is not a bigger bonus but a rebuild of the very rhythm in which the organization sets people tasks and responds to their effort—before the labor market responds.

The situation: the people leaving are not the ones anyone expected

The BANI world (Brittle, Anxious, Nonlinear, Incomprehensible—the name for the conditions in which almost any business operates today: old forecasting models can’t take a hit, and familiar solutions stop working as reliably as they used to) has made precisely this problem more acute than many others. From 2023 to 2026, the labor market for strong engineers never once cooled: demand for people able to design complex systems and find their way around AI-based tools consistently outstripped supply, and moving between companies came to take weeks rather than months. Against this backdrop, the global picture of employee engagement, according to the research company Gallup (its 2026 report, the annual State of the Global Workplace study), already looks alarming at the level of managers: the engagement of managers fell from 27% to 22% in a single year. In other words, managers’ very ability to spark their teams’ interest in the work is declining faster than you would think if you looked only at turnover figures for rank-and-file employees.

The company—let’s call it Orbit; it is a generalized, composite example of a mid-sized product IT company—has about four hundred employees, a hundred and fifty of them software engineers. The product is a platform for automating the internal processes of mid-sized businesses. The market is competitive but not hopeless: Orbit has a strong technical reputation, customers love the product, and the company has money for growth. On paper, everything is fine.

In the spring, the chief product officer notes something that can’t be written off as chance: over a year, seven of the company’s lead engineers have left—people from what the company called its “core,” not by title but in fact: these were the people everyone else went to for a solution when nothing else helped. Seven out of thirty-two lead specialists at this level—more than one in five. Meanwhile, turnover among rank-and-file developers stayed at the industry’s usual level, about twelve percent a year. It was precisely the best who were leaving, and they were leaving for competitors or startups for comparable or even slightly less money.

This is not some local oddity of Orbit’s—it is a stable pattern, well documented in the real market. The analytics firm SignalFire, which studied the career trajectories of engineers at leading AI labs by analyzing profiles on a professional social network, recorded a striking asymmetry in its 2025 State of Talent report: two-year retention at Anthropic was 80%—the best figure on the market—while at OpenAI it stood at 67%, although, by industry accounts, Anthropic did not pay more than its competitor. What’s more, engineers left OpenAI for Anthropic roughly eight times more often than the other way around. This is a direct, empirical refutation of the most convenient management myth—that retaining strong engineers comes down to the size of their salary. If money were the main factor, the flow would run in the opposite direction.

Broader data are just as telling. Researchers at the Massachusetts Institute of Technology (Sull, Sull and Zweig, 2022), who analyzed more than 34 million employee profiles and 1.4 million employer reviews on the specialized platform Glassdoor, found that a toxic corporate culture predicts an employee’s departure about ten times more strongly than the size of their compensation. ADP, the HR and payroll services company, in its annual People at Work report for 2026—a survey of more than thirty-nine thousand workers in thirty-six countries—found a similar pattern, expressed differently: only 22% of employees are confident that their position in the company is secure, and those who feel this confidence are six times more likely to be engaged in their work than those who don’t. Money is not absent from this picture altogether—it is a necessary condition, but far from a sufficient one and, more importantly for a manager, far from the scarcest.

The first reaction of Orbit’s leadership is predictable and typical of the industry. HR (out of old habit it is sometimes still called the personnel department, though human resources management is the more accurate name) surveys the people who are leaving and collects what are called “exit interviews”—a short conversation with an employee in their last week before leaving, to understand the reason. The answers are monotonous: “I found an offer with a higher salary,” “I want to try myself on problems of a different scale,” “I feel I’ve already done everything I could here.” The CFO proposes the obvious: raise salary bands for lead engineers by fifteen to twenty percent and introduce retention bonuses—a fixed payment awarded if the person stays for another year. The measure is introduced. The departures continue at the same pace.

This is where the real work begins: not asking “Why are they leaving?” but asking “At what point did their path inside the company stop working the way it used to?”

Diagnosis through the cycle: what exactly got stuck

To understand what broke, it helps to recall what a person’s productive work on a task consists of in the first place—not as a metaphor but as a sequence of quite physiological states, examined in detail in Part I of the book. First the person searches: looks around, assesses the task, estimates whether it is within their abilities and whether it is worth the effort. Then they make a decision—to take it on or not. Next comes the Action phase—the one in which the person produces a result—and it is accompanied not by joy but rather by working tension, focus, mild arousal. This is borne out by the research of the psychologist Eddie Harmon-Jones, who showed that a state resembling the thrill of the chase or even irritation is physiologically linked not to a wish to retreat but to movement toward a goal—it is this, and not complacency, that pushes a person to see difficult work through to the end. And finally, the Reward phase: the result has been achieved, it has been recognized, and the brain’s dopamine system—the one responsible for the feeling of “I did it”—registers that reality has matched or exceeded expectations. After that the tension subsides, and the person moves back into a state of search—but, importantly, at a new level of task, because they have grown. This is not a one-off loop but an expanding spiral: each subsequent cycle should offer a slightly more complex, slightly more meaningful task than the one before.

It is the neuroscience of reward—the research on the dopamine system conducted by the neurophysiologist Wolfram Schultz and his school—that supplies the key fact here: the brain responds not to a reward as such but to the difference between what a person expected and what they got; this is called “reward prediction error.” If the result exactly matched expectations, the dopamine burst is close to zero. If it exceeded them, the burst is strong. If the result is the same as last time, and the task was of the same difficulty as last time, the body registers “nothing has changed” and quickly goes back to looking for a new stimulus. Hence a practical conclusion that is often forgotten: a reward system can’t be “set up once and left alone”—it works only when the challenge grows along with the person’s skills.

When Orbit’s working group—the chief product officer, the head of HR and three engineering managers—walked through this logic and laid it over the stories of the seven engineers who had left, the picture came together quickly, and it turned out to be richer than the initial hypothesis that “competitors pay more.”

First, a gap emerged between the Action phase and the Reward phase. Orbit’s lead engineers had plenty of complex tasks—architectural decisions, working through hard technical debt, mentoring junior developers; there was no shortage of work at all. But recognition of that work was set up badly: bonuses and promotions were tied to the annual performance review cycle, once every twelve months. An engineer would solve a difficult problem in March but could hear anything about it beyond a verbal “thank you” at a meeting no earlier than December. By December the task itself was long forgotten—both the engineer and their manager were already busy with the next things. The reward came so late relative to the effort that the dopamine system, figuratively speaking, had already stopped waiting for it: the reward prediction error dropped to zero not because there was no reward but because the link between a specific effort and a specific response had disappeared. This is exactly what a longitudinal study by Gallup together with the employee recognition platform Workhuman shows (2024, nearly thirty-five hundred workers followed over time): high-quality, timely recognition reduces the likelihood that a person will leave within the next two years by 45%. The key word here is timely: recognition separated from the moment of effort by nine months works, physiologically, almost like no recognition at all.

Second, another gap emerged—between Search and Decision, that is, on the way into a task rather than on the way out of it. Orbit’s lead engineers had stopped choosing their own tasks. Earlier, at the company’s early growth stage, work had been distributed almost ad hoc: an engineer would see a problem, offer to solve it and get that piece of work. As the company grew, project management by the book appeared—from a process standpoint, this was a correct, necessary sign of maturing. But a side result was that tasks were now handed down from above, according to a plan drawn up a quarter in advance. The Search phase—the very one in which a person decides for themselves that a task is worth their effort—disappeared for the lead engineers altogether. All that remained was the phase of carrying out someone else’s decision. And research on the self-regulation of behavior by the psychologists Charles Carver and Michael Scheier shows that a subjective sense of control over the choice of a goal is not an embellishment but a necessary condition for sustained motivation. Without it, even a task with interesting content is perceived as imposed.

Third—and this came out only after direct conversations with three of the engineers who had left, who were persuaded to speak more frankly than a formal exit interview allows—there was a third, subtler gap. For these people, tasks inside Orbit had stopped growing in complexity. The company had reached a stage of product maturity at which most of the architecture was already built, the main technical problems were solved, and a lead engineer’s day-to-day work had shifted from creating something new to maintaining and fine-tuning what already existed. For an engineer who two or three years earlier had eagerly designed a system from scratch, this felt not like stability but like their own growth coming to a halt. The challenge curve had flattened, and in the cyclical logic of “Search—Decision—Action—Reward,” it is precisely the growing complexity of the task from one loop to the next that makes a turn of the spiral a turn rather than marking time. When the turns stop differing from one another, a person—subjectively, and quite accurately—stops growing, and this feeling outruns any figure in the employment contract.

There is also a fourth layer, which the working group found not in its own data but by comparing its situation with the broader industry context—and it is worth keeping in mind because it concerns not Orbit in particular but the whole industry as of mid-2026. LeadDev’s study of the state of engineering leadership (2025, more than six hundred engineering leaders and developers surveyed) documents a phenomenon known as “survivor syndrome”: after the waves of layoffs and restructuring that swept through the tech industry in previous years, what grows among the strong specialists who remain is not gratitude for keeping their jobs but hidden exhaustion—critical levels of burnout are found in one in five. Some of the engineers who left Orbit were coming from exactly this state: a few years earlier, the company had already been through a headcount reduction in another line of business, and although it did not directly affect the lead engineers, they did the work of their laid-off colleagues for almost a year before staffing was restored—and this added accumulated fatigue to the three gaps already described, even though no one ever explicitly named it out loud.

Putting these gaps together, the working group arrived at a diagnosis directly opposite to the original one: the problem was not pay. Pay was at market level, in places above it. The problem was that the whole “challenge–reward” rhythm, which at the company’s early stage had worked by itself, without anyone designing it, stopped working as the company accumulated processes—and no one deliberately rebuilt it for the new scale. Money in this picture is not a solution but an attempt to compensate with money for what should be compensated by the way the work is organized. Such compensation works exactly until the next offer from a competitor, because it changes nothing in the person’s day-to-day experience. This fits precisely with a McKinsey finding (its 2025 workforce monitoring): contrary to the popular narrative of a “search for meaning” as the main motive for changing jobs, the factor most often named as decisive for staying is not inspiration at all but a sense of security and clarity about one’s own position—that is, the very order in the “challenge–reward” rhythm, not a one-off emotional high.

There is also a broader, industry-wide backdrop that gives this diagnosis additional weight and explains why 2026 in particular made the problem more acute than previous years. The World Economic Forum’s Future of Jobs Report (2025) estimates that by 2030, 59 out of every 100 workers in the world will need reskilling to some extent, and about 11 of them, by the report’s calculations, are unlikely to get it in time—roughly 120 million people at risk of their own skills becoming obsolete. For the engineering profession, where AI-based technologies change the very nature of day-to-day work within a year or two rather than a decade, this is not an abstract figure somewhere in the future but the backdrop against which every decision to stay or leave is being made right now. A strong engineer in 2026 implicitly but constantly asks not only “Is this interesting for me here today?” but also “Will I still be an in-demand specialist in two or three years if I keep growing at this particular company?” A company that can’t answer this question clearly—because it doesn’t itself offer tasks of growing complexity—loses the competition for such people regardless of what the employment contract says about salary.

The solution, step by step

Orbit’s working group decided not to “retain” departing engineers one by one with counteroffers—a practice the company had used before and deliberately abandoned, because a counteroffer at the moment of resignation treats the symptom rather than the cause and almost always merely postpones the departure by a few months. Instead, it launched a rebuild of the rhythm itself for all hundred and fifty engineers at once. Here is what it consisted of.

Step one. Shortening the distance between action and recognition. The annual review was kept as the basis for salary decisions—that makes sense, since salary can’t be revised every week. But alongside it, a short feedback rhythm was introduced: once every two weeks, the direct manager and the engineer hold a fifteen-minute conversation, not about task status (there are other meetings for that) but specifically about what in the past two weeks’ work deserves recognition and why. The key condition is that recognition must be specific and tied to the substance of the solution rather than being general approval: not “well done, good work,” but “the way you cut the system’s response time by forty percent by changing the order of database calls—that’s exactly the level of solution that moves the product forward.” The difference seems small, but it is specificity that makes recognition just as real a reward, one distinguishable from the last—rather than background noise of approval that the brain quickly gets used to and stops responding to. This is a direct practical application of the Gallup and Workhuman finding on high-quality recognition: the company deliberately restructured recognition so that it would be timely, specific and different in form from the time before.

Step two. Giving engineers back the Search phase—that is, the right to choose their tasks. Out of the overall quarterly product development plan, a share of the team’s capacity was set aside—twenty percent of the lead engineers’ working time—for tasks that an engineer chooses from a list of technical and product problems agreed with product leadership, instead of receiving them by assignment. The list is put together jointly: product managers propose what matters from a business point of view, engineers propose what matters for the technical health of the system, and then the engineer chooses where to invest this twenty percent. This is not “time for personal projects” in the pure sense—it has to benefit the company—but the choice within the list belongs to the person. This restores precisely the part of the cycle responsible for the sense of control over one’s own work.

Step three. Introducing a deliberate increase in task complexity—instead of the spontaneous one that used to happen by itself in a growing market and has now stopped happening. Every six months, for each lead engineer, the manager and the engineer together set out not only a list of tasks but also what makes these tasks objectively more complex than those of the previous six months: a new subject area, a larger system, a higher cost of error, leading a larger group of people. If no such increase in complexity can be found, this is openly called what it is: a signal that the person needs either a new direction within the company, or a rotation to another team, or an honest conversation about the fact that growth here is temporarily limited. Previously, no one said anything about this until the person resigned; now it has become a subject of planned management work rather than a cause for awkwardness.

Step four. Rebuilding the bonus system itself—not making it bigger but changing how it works. Instead of a single annual bonus, two tracks were introduced: the annual track was kept for the basic market-based salary review—a hygiene minimum, not a motivating factor but a condition without which any conversation about engagement makes no sense at all (this matches the long-established idea in management psychology that pay below market level demotivates strongly, while pay above market level motivates weakly and briefly—the same logic confirmed by the gap between Anthropic and OpenAI: pay no higher than the competitor’s did not prevent better retention, because the base level was above the sufficiency threshold). And alongside it, a targeted track was added—small but immediate bonuses that a manager can award within one or two weeks after a task of exceptional complexity has been solved, without having to wait for the annual cycle. The amount of these bonuses is modest by the company’s standards, but speed—with the reward arriving in days rather than months—turned out to matter more than the amount: it closes the very feedback loop between effort and result that the psychologists Carver and Scheier wrote about, as applied to personal self-regulation, in their earliest work on the cyclical nature of human activity.

Step five. Introducing internal rotation for those whose personal turn of the spiral had flattened out. The engineers whose tasks were found to have stopped growing in complexity because of the product’s maturity were offered not the option to “stay and put up with it” but a deliberate move—to a new product line, to an internal project to bring AI-based tools into the company’s own development processes, or to a mentoring role for new employees, with their tasks formally revised for that role. Internal rotation became a recognized, encouraged path rather than an escape, which until then had tacitly been seen as almost a betrayal of the team. A separate item within the same step was work on “survivor syndrome”: for the engineers who had carried an extra workload during the earlier headcount reduction, the company introduced a direct, explicit conversation about accumulated fatigue and, where possible, a temporary reduction in workload without any loss of status or growth prospects—that is, a managerial acknowledgment that for these people the Action phase had been artificially stretched out and needed to be closed deliberately, instead of waiting for it to close by itself.

The five steps were introduced not all at once but over two quarters: first feedback and targeted bonuses, as the fastest and least costly change; then the right to choose tasks and complexity planning; and only last of all rotation, because it required open positions to be available inside the company.

The result

Below are illustrative, composite figures for Orbit, typical of companies of this size and profile; they do not describe any specific existing organization but show the order of magnitude achievable in similar situations over a year of systematic work. (The figures for real companies and studies cited earlier in this chapter—SignalFire, MIT Sloan, Gallup, Gallup × Workhuman, ADP, McKinsey, LeadDev, the World Economic Forum—are real and verifiable; they should not be confused with the indicators of the fictional Orbit.)

Twelve months after the rebuild of the rhythm began, attrition among Orbit’s lead engineers fell from almost twenty-two percent a year (seven out of thirty-two) to six or seven percent—that is, to a level that can be considered healthy natural renewal of the team rather than a warning sign. The average time between solving a difficult problem and receiving recognition for it (formal feedback or a bonus) shrank from several months to one or two weeks. The share of lead engineers who, in the engagement survey, agreed with the statement “I choose what to work on for a significant part of my time” rose from roughly fifteen to almost sixty percent. The number of internal moves between teams and projects—something that used to happen once every year or two and was seen as an exception—grew several-fold and became normal practice, which further reduced external attrition, because some of those who used to look for new complexity elsewhere began to find it inside.

One more point deserves a note of its own: the whole program cost the company less than the originally proposed across-the-board raise of fifteen to twenty percent for all lead engineers: targeted bonuses tied to real results came out cheaper than a flat raise for everyone without distinction, including those whose motivation had nothing to do with money at all. In other words, the company not only retained people better—it did so while spending less than it had planned to spend on a solution that would not have worked. This is the same conclusion the MIT Sloan study reaches on an incomparably larger body of data: trying to retain people with money alone, amid an unhealthy culture and a broken work rhythm, is a poor return on spending, because it targets the wrong cause.

An important caveat, typical of stories like this: some of the seven engineers who left would have left anyway—some for personal reasons, some because they had genuinely exhausted what this particular company could offer at this stage of its development—and that is normal; not every departure is a management failure. The aim of rebuilding the rhythm is not to keep absolutely everyone but to make departures the result of a person’s conscious choice, not the result of the company itself, without meaning to, breaking the mechanism that makes work interesting. The gap between Anthropic and OpenAI mentioned above is telling in this respect too: even the market leader in retention doesn’t keep a hundred percent of its people—it keeps enough for its two-year retention to be noticeably higher than that of its closest competitor, and that is already a strategic advantage, not absolute invulnerability.

The engagement of team leaders—those very line engineering managers who held the fifteen-minute recognition conversations every two weeks—deserves a word of its own. The conversations turned out to be not only a tool for engineers but also an unexpected remedy for the managers themselves, whose engagement is under pressure too, if you recall the Gallup data cited at the beginning of the chapter on managers’ engagement falling from 27% to 22% on average worldwide. A regular, specific conversation about what exactly a team member did and why it matters forces a manager to reformulate for themselves, every two weeks, the meaning of the team’s work—and Orbit separately recorded that over the same year the engagement of the engineering managers themselves rose more noticeably than that of their reports, although this was never a goal of the program. The side benefit turned out to be no less valuable than the main one: a rhythm of recognition works in both directions—it supports not only those who are recognized but also those who do the recognizing.

When the information a decision requires is contradictory and incomplete—and in logistics it is almost always contradictory and incomplete—the right response from a leader is not to wait until the data converge into a single, consistent picture. That picture will never appear, because the market changes faster than the gap in reporting can close. The right response lies in a protocol: a procedure agreed on in advance that forces the team to make a decision on a fixed, knowingly incomplete amount of information—and in rethinking the fear of making a mistake itself. Fear is not a signal to “wait a little longer” but a selection tool: it shows precisely which decisions need to be kept reversible and which can be made at once and for good. Those who confuse these two things either drown in analysis that never ends or take risks where no risk is needed. Those who tell them apart make decisions faster than their competitors and don’t pay for that speed with quality. What follows is the story of how this looks in practice at a company that lives by cargo, containers and schedules, where a one-day mistake turns into money faster than in any other industry.

The situation: three reports—three different realities

A mid-sized logistics operator—tens of thousands of shipments a month, its own truck fleet, partnerships with sea and rail carriers, warehouses in several regions—faces a problem familiar to anyone who has ever managed a supply network: one of its key transport corridors, which carries a noticeable share of the company’s revenue, suddenly becomes unreliable. Formally, nothing catastrophic has happened: there has been no accident and no declared force majeure. There is only a growing sense that the corridor will soon start to fail, and three sources of data that neither confirm nor refute that sense but argue with one another.

The operations report—dry figures from the warehouses and the dispatch center—shows that the average transit time for cargo along the corridor has risen by 12% over the last quarter but remains within a range that has occurred before in peak season. By these figures it is too early to worry: the fluctuation is within the system’s normal noise. The finance report says something else: the cost of insuring cargo on this route has nearly doubled over the same quarter, and two major insurers have already refused to take on new contracts for this corridor without an additional risk premium—the capital market, which by definition sees risks earlier than operating statistics do, is clearly pricing in something that is not yet visible in the delivery metrics. And the third source—the security department and the regional representatives—brings to the table the material that is hardest to formalize: growing unofficial signals of political instability at a key junction of the corridor, scattered reports from partners about delays at the border with no clear explanation, rumors of impending sanctions restrictions that no one can confirm with documents but that keep recurring from mutually independent sources often enough that they can’t be written off as chance.

The three reports, laid side by side on the CEO’s desk, do not add up to a single picture. The operating statistics say “wait and watch.” The financial market says “the risk is already material and expensive.” The field intelligence says “something is happening that we can’t yet name precisely.” None of the three sources is lying or mistaken within its own area—each conscientiously reports what it sees with its own measuring tools. The problem is not the quality of the data but the fact that three reliable slices of reality, taken at different speeds and with different degrees of formalization, are inevitably out of sync in time: prices in the insurance market react to a risk weeks before it shows up in operating metrics, and rumors from the ground often run ahead of both but can’t be verified until it is too late.

The company’s board of directors reacts the way almost any board of directors reacts at a point like this: it demands more complete information before deciding the fate of the corridor—whether to cut the volume of shipments through it, whether to look for an alternative route in advance, given that the alternative is more expensive and slower, or whether to wait for a clearer picture. A working group is formed. An in-depth analysis is commissioned from outside consultants. A series of meetings is held at which each side—the operations division, the finance department, the security department—methodically defends its own version of reality, because each conscientiously relies on its own data, which the other departments can’t see. Six weeks go by. The consultants’ report arrives—and instead of resolving the contradiction, it adds a fourth voice: the outside analysts put the probability of a serious disruption on the corridor within six months at “15 to 45%”—a range that is formally honest and practically useless for making a decision, because its lower and upper bounds call for diametrically opposite actions.

Over those six weeks, nothing stays the same. Some competitors, who are not in the habit of waiting for all the data to be reconciled into a single consistent table, have already begun redirecting cargo to alternative routes—more expensive, but more predictable. The company’s customers, who follow the news around the region themselves, start asking direct questions about the reliability of deliveries and get evasive answers, because the company itself does not yet have a decision it could honestly announce. Some of the contracts for the coming quarter go to the carriers that have already settled on a route—not because their route is objectively better, but because they have an answer, while our company has only a process for working one out. And in the seventh week the corridor really does begin to fail: first scattered delays at the border, then one truly serious incident with detained cargo that costs the company an amount comparable to what it had saved over six months by not using the alternative, more expensive route.

The telling detail of this story is that in hindsight, after the incident, none of the three departments can say it was wrong. The operating statistics really did show a moderate rise in the indicators, not a collapse. The insurance market really did assess the risk correctly in advance. The field signals really did foreshadow exactly what happened. This company’s problem was not the quality of any one of the three data sources taken on its own. The problem was the absence of a procedure that would allow a decision to be made without waiting until three streams of information—contradicting one another, yet each accurate in its own way—converged into a single, consistent picture, because for decisions of this kind they will most likely never converge in time.

Diagnosis through the cycle: what exactly got stuck

In the logic of the emotional cycle analyzed in Part I of the book, a person and an organization move through a sequence of several phases: Search (orienting oneself in a new, unclear situation, often accompanied by anxiety), Decision (the transition from uncertainty to a chosen direction of action), Action (the productive tension of working toward a goal) and Reward (closing the cycle and recognizing the result, after which comes Search again, but now at a new level of task complexity). This mechanism has a physiological basis: the amygdala—a brain structure that, according to the neuroscientist Joseph LeDoux, assesses potential threat quickly and before conscious, slow reasoning kicks in—triggers a state of anxiety in response to contradictory signals long before reason manages to bring them together. Moderate anxiety at this point is useful: it sharpens attention and makes people check sources and look for alternative explanations. But, as the psychologist Jeffrey Gray showed, once a certain threshold is crossed, a different system switches on—behavioral inhibition—and then a person or an organization does not just warily search for an answer but physically stops, freezing in anticipation of more complete information that will remove the need to choose.

The logistics company’s board of directors got stuck at exactly this point—in the Search phase, stretched beyond any reasonable limit—for two overlapping reasons.

The first reason is the illusion that a decision can and should be put off until the data stop contradicting one another. This is not an idle managerial mistake but a perfectly understandable human reaction: the brain perceives inconsistent data as higher uncertainty than data that are simply scarce—and high uncertainty, according to the appraisal theory of emotion developed by the psychologist Richard Lazarus, is interpreted as a heightened threat, because it is unclear what resources are needed to cope with it. A board of directors that receives three mutually exclusive reports intuitively decides that the task is to reduce them to one before acting. But there is a body of management research by Herbert Simon—an economist and psychologist who won the 1978 Nobel Prize in Economics for his research into the decision-making process in organizations—that describes this trap directly. Simon showed that under a real shortage of time and information, managers are physically incapable of maximizing a decision by going through every alternative and waiting for complete data—they rationally “satisfice,” settling for the first solution that is acceptable on the data available here and now. This is neither weakness nor carelessness but the only rational strategy in a world where data are always incomplete and always will be: waiting for them to be complete means postponing the decision indefinitely, because completeness will never arrive before the situation changes once again.

The second reason runs deeper and is more dangerous than the first: the board did not distinguish between different types of decisions within a single problem and demanded the same completeness of data for all of them at once. The question “should we already partially redirect cargo volume to the alternative route now?” is a reversible decision: if the alarm turns out to be false, part of the volume can be moved back to the main corridor without catastrophic losses. The question “should we terminate long-term contracts with partners on the main corridor?” is an almost irreversible decision: broken relationships with local partners can’t be quickly restored, even if the situation in the region stabilizes a month later. The board intuitively demanded the same certainty for both types of decisions—and in the end made neither of them in time, even though the first could have been made almost immediately, while the second really was worth postponing. Lumping reversible and irreversible decisions together into a single undifferentiated mass of “strategic choice” is a typical mechanism by which an organization loses weeks where it could be losing hours.

The fear itself deserves a separate look—the emotion felt in this situation both by the rank-and-file dispatch staff who noticed the first disruptions and by the members of the board watching the cost of insurance climb. The psychologist Nico Frijda described emotions as states of action readiness: fear is not just an unpleasant experience but the body’s physiological preparation for specific behavior—to retreat, to freeze or to flee. There is sound sense in this readiness to freeze: caution in the face of a real threat that is not yet fully understood is an adaptive reaction, not a pathology. The board’s mistake was not that it felt anxiety in the face of contradictory data, but that it used this anxiety the wrong way—as a reason to put off any action at all—instead of using it as a filter that separates the decisions that genuinely require caution and greater certainty from the decisions that only appear to justify that anxiety because they involve money and headlines, but are in fact easily reversible.

Here it is worth citing an example of a different kind of thinking about the very same class of problems—the research of the psychologist Gary Klein, who spent several years studying how experienced fire commanders make decisions under a real shortage of time and information: 26 commanders, 156 decisions analyzed in smoky, fast-changing conditions where delay costs lives. The result was unexpected for classical decision theory: in the overwhelming majority of cases—more than 80% of the episodes analyzed—the experienced commanders did not compare several courses of action with one another, weighing the pros and cons of each. They recognized the situation as a familiar pattern, immediately saw one plausible solution and mentally checked whether it would work—and if it did, they acted, without spending time building a full set of alternatives. This model became known as “recognition-primed decision” making. The logistics company in this story tried to do exactly the opposite: instead of relying on the experience of the dispatchers and regional representatives who had watched similar situations on other corridors for years, it pushed the question up to the level of an abstract comparison of probabilities—and an abstract comparison of probabilities, as it turned out, does not resolve the contradiction between different data sources but merely records it as a wide range that is useless for action.

The conclusion of the diagnosis is this: the organization got stuck not because data were scarce—on the contrary, data were plentiful, coming from three independent and generally reliable sources. It got stuck because there was no protocol that determined in advance at what minimum, knowingly incomplete amount of information a decision must already be made, and no distinction between decisions that can be made at once because they are easy to reverse and decisions that really are worth holding back because the cost of a mistake is irreversible. The fear the board felt was not an error of perception—the risk signal was real and, as later events showed, essentially correct. The error was that this signal was used as a universal brake instead of a precise tool for sorting decisions.

The solution: a protocol instead of waiting for complete data

The turning point in this story came not after the incident with the detained cargo—although that was the painful occasion for revisiting the practice—but about two months later, when the company’s new chief operating officer (hired partly to solve precisely this problem) proposed not a one-off decision on this particular corridor but a standing protocol for all future situations of this kind. Below is the sequence of steps that formed the basis of the protocol.

Step one. Separate decisions by reversibility before separating them by content. The first action was not a deeper analysis of the data but a simple management procedure: every major decision related to the corridor was broken down into sub-items, and for each one it was stated explicitly whether it was reversible or not—and if it was, what a rollback would cost. Redirecting 20% of cargo volume to the alternative route: reversible, with a moderate rollback cost. Expanding the capacity of the alternative route beyond current needs: partially reversible, with losses on unused capacity. Terminating long-term contracts with partners on the main corridor: practically irreversible. It was this separation, not more complete data, that turned out to be the main missing link: as soon as the board saw that some of the decisions could be made almost immediately, because they were easy to undo if the alarm proved false, the paralysis that had gripped the whole package of questions lifted from most of them.

Step two. Introduce an information sufficiency rule—with a concrete threshold, not an abstract feeling. The company adopted a management rule inspired by the publicly stated practice of Amazon founder Jeff Bezos, set out in his 2016 letter to shareholders: most decisions should be made with roughly 70% of the information you would like to have, not 90%, because waiting for ninety percent almost always means the decision is made too slowly—at that level of uncertainty the cost of delay is usually higher than the cost of a moderate risk of being wrong and then correcting course. For reversible decisions, the company turned this principle into a concrete threshold: if at least two of the three main data sources—operating statistics, the market’s financial signals and field intelligence—point in the same direction, a decision is made within one calendar week, without waiting for the third to come into line with the others. For irreversible decisions the threshold was higher—agreement of all three sources or a clear breach of a damage threshold agreed on in advance—but here, too, a maximum review period was set so that the Search phase could not drag on without limit.

Step three. Make fear a legitimate selection criterion, not a reason for silence. This step directly addressed what the diagnosis had called the misuse of anxiety. The company introduced a simple management requirement for anyone putting forward a “let’s wait” argument: the argument is considered only together with an answer to the question, “If we wait and our worst fears come true, what price will we pay for the delay—and can today’s inaction be rolled back as easily as an action could be?” If the answer showed that inaction was in itself irreversible or costly—and that was exactly the case with the delay on the corridor—then anxiety stopped being an argument for waiting and became, on the contrary, an argument for immediate partial action. The phrase that took root in the company after this story and is still repeated at meetings: “Fear is not a reason to stand still; it’s a hint about where exactly you need to be careful.” In practice the difference looked like this: where a decision was reversible, anxiety no longer blocked action—the company acted while keeping the right to change its mind. Where a decision was irreversible, the same anxiety had every right to demand additional checks, and no one called that caution procrastination.

Step four. Give field data a voice on a par with formal reporting. One of the hidden sources of delay was a matter of status: reports from the regional representatives and dispatchers, who stood closest to the real situation on the corridor, carried less weight with the board than the figures in the operations and finance reports, because they were not presented as tables and charts. This directly contradicts what Klein’s research on fire commanders showed: an experienced person’s recognition of a familiar pattern on the scene is often more accurate than formal analysis simply because it carries information that cannot yet be measured and put into a table. The company introduced a rule: at any meeting that considers a decision on a high-risk corridor, the field representative speaks not last, as before—after the figures have effectively set the direction of the conversation—but first, before the metrics reports frame the discussion. This small change in the speaking order turned out to matter for management: it kept the formal figures from discounting, right from the start, the less formalized but often earlier warning signal from the ground.

Step five. Put the protocol in writing and test it on a lower-risk case before relying on it the next time. The company did not stop at a review of the past incident—it formalized the four previous steps into a short document that fits on a single page: a protocol for deciding on incomplete data, mandatory for any similar fork in the road. The protocol’s key condition is a mandatory decision deadline (one week for reversible decisions, a maximum of four for potentially irreversible ones, with an interim review after two), fixed organizationally rather than left to the discretion of the moment. Six months later the protocol was deliberately tested on a lower-risk trial situation—a fluctuation in rates on another, less critical route—so that the board would gain experience applying the procedure before it again had to decide a question with a high cost of error. This is a direct application of the logic of self-regulation that the psychologists Charles Carver and Michael Scheier describe as movement toward a goal through successive, manageable steps with feedback: an organization, like a person, gets through the Decision phase better if it already has a practiced route of action, tested on lower-risk material, rather than an abstract instruction applied for the first time under maximum pressure.

The order of these five steps is not accidental. Separating decisions by reversibility gave the team the ability to act partially without waiting for all the uncertainty to be removed. The data sufficiency threshold turned the abstract demand to “act faster” into a concrete, measurable procedure. Redefining the role of fear freed the team’s hands where anxiety had previously blocked every move indiscriminately. Changing the order of voices at meetings gave weight back to a source of signals that was undervalued but often the earliest. And finally, putting the protocol in writing, with a trial run on a lower-risk case, turned a one-off management conclusion into a permanently working tool of the organization—one that outlives changes in the specific people sitting at the board table.

It is also worth spelling out what this protocol fundamentally is not. It is not a call to act rashly or to neglect analysis—the very logic of the protocol requires explicit, disciplined analysis of the incoming data, just within a deadline agreed on in advance rather than endlessly. Nor is it a universal recipe of “always act fast”—quite the opposite: the protocol is specifically designed to serve two opposite cases equally well. Where delay is expensive and a mistake is easy to fix, act fast; where a mistake cannot be fixed, deliberately hold the decision back—but with a clearly marked deadline and a list of exactly what needs to become clear within it, not with a vague “let’s wait a little longer.”

A similar logic of the cost of delay in a networked, tightly coupled economy was vividly demonstrated by a case that has nothing to do with our company’s story but illustrates the same principle on a global scale: the accident of the container ship Ever Given, which blocked the Suez Canal in 2021. Each day that one ship stood still at one narrow point in the world’s logistics network held up, by the estimate of the maritime publication Lloyd’s List, about $9.6 billion in cargo—some $55–60 billion over six days. The six-day difference between a fast and a slow resolution of a local blockage turned into a difference of tens of billions of dollars for thousands of companies that had nothing to do with the accident—a precise illustration of why in logistics, where events at one point of the network instantly spread along the entire supply chain, the cost of delay in making a decision is often higher than the cost of a mistake in the decision itself.

The result

The figures that follow are an illustrative, composite example typical of mid-sized logistics operators with a diversified network of corridors, not the indicators of any specific real company.

A year after the protocol was introduced, the average time from the moment contradictory signals about a corridor first reached management’s desk to the moment a decision was made on reversible questions had fallen from the previous five to seven weeks to four to six days. Decisions made under the “two of three sources agree” rule accounted for about 70% of all decisions on risky corridors over the year—that is, in most cases the company really did have enough incomplete, partly contradictory data to act on, and further waiting added no quality to the decision but merely delayed it. The average financial cost of delays and forced unscheduled measures on risky routes fell by roughly 35% on an annual basis compared with the previous period—primarily because the partial redirection of cargo volume now happened in advance rather than after a disruption had already occurred. The share of customer contracts lost because competitors settled on a route faster fell from its peak to almost zero on the routes where the protocol was applied systematically.

What did not happen deserves a note of its own—and it is no less important a result than what did. The fear of some board members that speeding up decisions while data remained incomplete would lead to more wrong decisions and more losses from them was not borne out: the share of decisions later recognized as mistaken and requiring a rollback stayed at its previous level—about 12% of all decisions on reversible questions. In other words, the company did not start making worse decisions; it started making them faster with the same level of accuracy. This is directly in line with an observation recorded in research on managerial decision-making: organizations where decisions are made quickly are almost twice as likely to also show high decision quality—speed and thoroughness turn out to be not opposites but companions wherever there is a disciplined procedure rather than mere impatience.

The fate of the very corridor that started the whole review deserves special emphasis. The concerns of the regional representatives and the rise in insurance costs turned out to be justified: the situation at the key junction of the corridor only grew more serious over the next year and a half, and the company, having managed to move its main volume to the alternative route—first partially, then fully—avoided the level of losses suffered by several competitors who kept waiting for complete clarity. This does not mean that intuitive anxiety is always right—sometimes the protocol leads to a reversible action that then has to be rolled back because the alarm was not borne out, and that, too, is a normal, expected result of a system built on making decisions with incomplete data rather than on guessing the future.

There is a managerial temptation to read this story as a story about courage—as if management simply needed to be less afraid and quicker to make up its mind. That is an imprecise reading that misses the main point: what works is not giving up fear but precisely distinguishing where fear is a signal of real irreversibility that demands caution and where it is background noise before a decision that is easy to undo if it turns out to be wrong. The company did not become braver in the sense of being willing to take more risk. It became more precise about where a risk is genuinely worth taking on immediately and where it is worth resisting and waiting—but with a clear deadline for the wait, not indefinitely.

When a board of directors loses trust in a company’s management, the first instinct is to replace people: fire the CFO, change the CEO, find someone to blame and show investors that the culprit has been punished. Sometimes this is necessary, but it almost never solves the problem itself, because the problem rarely lies with a specific person. It lies in the fact that the board, as a collective body, is stuck in a single emotional phase—anxiety that seeks not a solution but immediate release—and in that state it is physically incapable of calmly considering any complex strategy, however many times it is presented anew. Managing such a board means not persuading it harder but deliberately taking the meeting through the same sequence of states that a single person goes through when solving a hard problem: first let the anxiety play out and don’t get in its way, then translate it into a precise assessment of risk, then let the board make a decision that it itself, collectively, recognizes as its own, and only after that lock in a small but real success that will restore trust in management. And—this is a separate, independent step—match the type of response to the type of damaged trust: a failure of competence is not treated with the same remedy as a suspicion of dishonesty. Skip any of these steps and you get a vote that looks like agreement but a month later turns into a fresh explosion of distrust.

The situation

The company—let’s call it Meridian—is a generalized, composite example of a vertically integrated oil and gas company of mid-range size for the industry: production, refining, some distribution assets, publicly traded shares and a board of directors of eleven people, five of whom are independent directors representing the interests of institutional investors. Autumn 2025: over three months the world oil price falls by almost a third amid a sharp slowdown in industrial demand and a simultaneous glut of supply; an accident at one of a competitor’s refineries leads to tighter industry regulation across the entire market; and Meridian’s own major investment project—the construction of a new refining complex, into which about forty billion rubles have already been invested—is eight months behind schedule and twenty-two percent over its original budget.

Taken separately, any one of these three events is something the board would have been able to discuss calmly. Together they create what this book calls a fragmented, contradictory picture: some figures say “stop the project, it won’t pay off at the current oil price,” others say “now, precisely when prices are falling, is the time to finish building refining capacity so as not to depend on the commodity cycle,” and still others say “the regulator could tighten requirements any day now to the point where the entire project would have to be redone.” At the November 2025 meeting, where the CEO is supposed to present a revised plan for the complex, something happens that people at the company will later call “that meeting”: a discussion scheduled for two hours runs for four and a half, voices are raised, one of the independent directors declares outright, in front of everyone, that management has been concealing the true extent of the delay since the summer, and the CFO, in response, goes on the defensive and starts recalculating figures on the fly, which looks like improvisation rather than a prepared position. The meeting ends without a decision: the board approves neither continuing the project nor stopping it, and instead instructs management to “present additional analysis”—a wording that in practice means the decision has been postponed while the anxiety stays right where it was.

What happens next is typical of boards stuck in anxiety: the anxiety does not go away on its own, because its cause—uncertainty about the oil price, about regulation, about the project itself—has not gone anywhere, while the decision has been postponed. Over the next six weeks, rumors of disagreements within management reach the board; one of the institutional investors represented on the board publicly voices doubts, at an industry conference, about Meridian’s ability to see the project through; and over this period the company’s shares lose about twelve percent of their value against a drop of roughly five percent for the sector as a whole—that is, the company loses noticeably more than the market precisely because of the distrust, and not only because of the general situation in the industry. By the December meeting, the board convenes not so much to discuss the project as to discuss whether management can be trusted at all.

It is telling how communication within the board itself behaved over those six weeks—not at the meetings but between them. Previously, board members had called the chairman or one another as needed, without any particular system; after the November meeting, the frequency of such calls increased severalfold, but their content changed: instead of discussing the substance of the project, they exchanged fears and speculation about what else management might be hiding. The chairman later admitted to a consultant that it was at this moment that he first physically sensed he was dealing not with working disagreements over a specific issue but with something more diffuse, in the background—an atmosphere in which any new figure from management was met with distrust in advance, before anyone had even had time to read it and think it over. The CFO, preparing materials for the December meeting, reported candidly that he could not tell what volume of analysis would be sufficient: since November the board had repeatedly changed the set of data it requested, and each new version of the report was perceived not as an answer to the request but as an occasion for a new round of questions—a characteristic symptom of anxiety that, having been given no explicit outlet, seeks one in an endless demand for new information, even though the real cause of the anxiety runs deeper than a lack of data.

It’s useful to step outside this particular company’s story for a moment and recall what happens in real life to trust in energy companies when crisis communication fails truly badly. After the accident at the Fukushima Daiichi nuclear power plant in March 2011, the Japanese government and the plant’s operator—TEPCO (Tokyo Electric Power Company)—informed the public about the scale of what was happening extremely reluctantly and inconsistently: there were no crisis communication protocols prepared in advance, and risk assessments were systematically understated. The result can be traced year by year: according to the annual poll by the Japan Atomic Energy Relations Organization (JAERO), the share of Japanese people who considered nuclear power necessary for the country fell from 35.4% in 2010 to 17.9% in 2017, while the share of those who distrusted it rose from 10.2% to 30.2% over the same period—seven years of steadily declining trust by an entire country in an entire industry, beginning with a single crisis period in which promises were made that were not backed up by facts. Meridian, on its own incomparably smaller scale, was risking exactly the same mechanism: not a one-off blow to its reputation but the start of a long, self-sustaining decline in trust, had the December meeting gone the same way as the November one.

Diagnosis through the cycle: what exactly got stuck

Before transferring this logic to a group of eleven people, it is useful to recall what happens to a single person facing a complex situation that threatens their goals. The psychologist Richard Lazarus, in his cognitive appraisal theory of emotion, showed that an emotion does not arise from a fact by itself—it arises from appraisal: an assessment of how significant the situation is for the person’s goals and whether they have the resources to cope with it. Klaus Scherer refined this idea: appraisal is not a one-time act but a sequence of quick checks—how novel the situation is, how important, how controllable, how consistent with what the person was already striving for. A board of directors that learns of an eight-month project delay and a twenty-two percent budget overrun goes through exactly the same sequence of checks—just collectively rather than individually: the situation is new (no delay on this scale had been reported before), significant (forty billion rubles and the board’s reputation with investors are at stake), barely controllable (the board can directly influence neither the oil price nor the regulator’s decision) and clearly at odds with what the board expected to hear back in the summer. On all four parameters at once, it is a textbook example of a situation that triggers not calm analysis but anxiety.

The neuroscientist Joseph LeDoux showed that the amygdala—the brain structure responsible for rapid threat assessment, before conscious deliberation even comes into play—reacts to signs of risk almost instantly, long before slow, considered reasoning kicks in. In an individual, this shows up as a surge of anxiety at bad news. In a group of people making a decision together, something similar happens, but with an important difference: one board member’s anxiety, voiced sharply out loud, proves contagious for the others—this is not a metaphor but what research on group behavior calls emotional contagion: the state of one person in a group is transmitted to others quickly, within minutes, through tone of voice, facial expressions and wording. That is exactly what happened at the November meeting: one independent director’s sharp claim that management had been hiding information instantly shifted the whole board’s anxiety from a background, manageable state into an acute one, in which asking calm clarifying questions became socially awkward—any question like “how great was the uncertainty back in the summer?” no longer looked like a search for facts but like an attempt to protect management.

Jeffrey Gray, the author of the theory of behavioral inhibition, explains the physiological basis of what happened next: when perceived risk and punishment are high, the behavioral inhibition system literally halts action—the organism (and, at the group level, the collective decision-making process) freezes instead of moving toward the goal. That is a precise description of how the November meeting ended: the board made neither of the two possible decisions—neither to continue the project nor to stop it—but chose a third, a pseudo-decision: “present additional analysis.” It is a classic symptom of getting stuck in the phase that the cycle of emotional regulation calls the Search phase, stretched beyond all measure: instead of assessing the risk and moving on to a decision, the board remained in information-gathering mode, because at that moment a decision seemed more dangerous than its absence. The problem is that the absence of a decision is also a decision, just the most expensive one possible: in six weeks of inaction the company lost twelve percent of its share value on the market—anxiety left without an outlet into a decision did not dissipate on its own but built up and gathered rumors around it.

To understand more precisely what exactly broke between Meridian’s board and its management, it helps to use a model of trust well established in organizational behavior research, proposed by Roger Mayer, James Davis and David Schoorman: one party’s trust in another rests on three perceived pillars—competence (whether the other party is capable of delivering what it promises), benevolence (whether it cares about the trusting party’s interests and not only its own) and integrity (whether its words and principles match its actual actions). The model matters because the three pillars are not interchangeable: a failure in one of them does not automatically mean a failure in the others, and different kinds of damage require different treatment. This is exactly what a later study by Peter Kim, Donald Ferrin, Cecily Cooper and Kurt Dirks confirmed: when the cause of distrust is a failure of competence (“you failed at the task”), trust is repaired more effectively through an open apology and acknowledgment of the mistake; but when the cause is an accusation of dishonesty (“you were deceiving us”), an apology works less well, and the more effective strategy is a firm denial of guilt followed by documentary proof of innocence, because an apology in response to an accusation of deception is subconsciously perceived as an indirect admission that the accuser is right.

For Meridian, this distinction proved decisive for the diagnosis. At the November meeting, the independent director had in effect leveled two different accusations against management at once, fused into one: first, that management had failed at the project (competence), and second, that management had concealed the extent of the delay (integrity). The CFO responded to both accusations in the same way—by getting defensive and recalculating the figures—which was an adequate response to the first accusation and a categorically inadequate one to the second: to an observer, a defensive, nervous reaction to an accusation of dishonesty looks exactly like the behavior of someone who really does have something to hide, even if in fact there is nothing to hide. The board, not distinguishing between these two types of damage to trust, reacted to both in the same way—with growing general distrust—because no one in the room stopped to explicitly separate the question “we are unhappy with the result” from the question “we suspect deception.”

One point that Meridian’s management misread in November deserves a note of its own: the reaction of the CFO, who began defending himself and recalculating figures on the fly. From the point of view of the psychologist Nico Frijda, who studied emotions as readiness for a particular action rather than mere experience, a defensive reaction is not a sign of incompetence but a precise physiological response from a person who perceives the board’s questions as an attack rather than as part of joint work on risk. The mistake was not that the CFO got nervous—almost no one is capable of reacting differently in such a situation—but that the meeting was set up in such a way that the board’s anxiety met management’s defensiveness head-on, without a buffer that would have let the anxiety first play out and name itself before moving on to the numbers.

The role of the chairman of the board deserves a word of its own here, because this is the level at which a management error most often occurs that makes the board more stuck rather than less. In most cases a chairman sees the role as that of an agenda moderator: keeping to the rules of procedure, giving the floor, summing up the votes. The psychologist Barbara Fredrickson, who studied how positive and negative states affect the range of actions available to a person, showed that a narrow set of responses, constricted by anxiety, is a direct physiological consequence of the anxious state, in which attention automatically narrows to the threat and its source. In November, Meridian’s chairman formally followed procedure impeccably—giving the floor in turn, watching the time—but did not notice that the very structure of the meeting gave the board’s anxiety-narrowed attention no way out other than to look for someone to blame. Procedure was followed, but the function a chairman is obliged to perform in a moment of collective anxiety—not merely running the meeting but deliberately managing the order in which the group goes through the assessment of risk—went unperformed, because no one had ever before formulated it as part of his role.

There is also a well-documented example of how far the restoration of trust can have to go when a crisis is truly deep. In 2004, Royal Dutch Shell was forced to sharply revise downward its stated proved oil and gas reserves—a figure that investors had for years relied on as the basis for valuing the company. The crisis of trust affected not only management but the very architecture of corporate governance: the separate structures of Royal Dutch and Shell Transport, which had historically run the company through a complex system of two boards of directors, were merged into a single company with one unitary board—the company acknowledged that its previous governance structure had itself been part of the problem, not just the specific people who had overstated the reserve figures. The lesson for Meridian here lies not in the scale but in the logic: restoring investor trust requires not just a new report with corrected figures but a visible, structural change in how the company makes and discloses decisions—that is, proof rather than statements.

This analysis produced a precise diagnosis, which the chairman of the board and an invited corporate governance consultant began working with in December: what had got stuck in Meridian’s board was specifically the link between anxiety and decision, and on top of that, two kinds of damage to trust that were different in nature—damage to trust in management’s competence and in its integrity—had been fused into one, although they required different treatment. The anxiety was entirely justified—the project’s risks were real, the delay was real, the fall in the oil price was real. The problem was not that the board was anxious but that the meeting was set up as if the anxiety did not exist: the agenda assumed an immediate move to a vote on the project’s fate, bypassing the stage at which the anxiety could have been voiced openly, heard and translated into specific, measurable risk criteria. Without that stage, the anxiety spilled out spontaneously, in the form of personal accusations, and blocked any sober decision.

The third part of the diagnosis, which the consultant formulated separately, concerned the CEO himself—and it came as a surprise to the board. The problem lay neither in his competence nor in his integrity: a review by an independent legal and audit team found not a single case of deliberate misrepresentation of data on the project’s progress—only a reporting system set up in such a way that bad news reached the board later than good news, because management, like any person in a state of uncertainty, kept putting off reporting a problem in the hope of solving it on its own in time. The higher the perceived cost of admitting a problem, the stronger the temptation to postpone that admission until the moment when the problem either resolves itself or becomes too visible to hide. The board interpreted this delay as malicious intent—that is, as damage to integrity—whereas in fact it was dealing with a systemic failure of competence in the way reporting itself was organized. That is why the solution lay not in a change of personnel but in rebuilding the system so that reporting bad news would become safer and faster than hiding it.

The solution, step by step

Meridian’s chairman of the board and CEO decided not to hold yet another meeting “as usual” in December, with the same structure that had already failed twice, but to deliberately rebuild the meeting process itself—and the period between meetings connected with it—around the following sequence: first let the anxiety play out, then move to the risk assessment, then to the decision, then to a small confirmed result, with an explicit distinction as to which type of trust each specific action restores. Here is what this rebuilding consisted of.

Step one. Setting aside separate time to name the anxiety—before moving on to the numbers. The December meeting opened not with a management presentation but with a twenty-minute round in which the chairman directly asked each of the eleven board members in turn, without objections and without discussion, to name in one or two sentences what worried them most about the situation with the project—not an argument, not a proposal, but the anxiety as such. The format specifically ruled out any response from management at this stage: the goal was not to solve the problem in twenty minutes but to let the anxiety come out in words rather than in hidden irritation, which otherwise breaks through later in the form of personal accusations. This technique draws directly on Amy Edmondson’s many years of research on psychological safety in teams: people and groups that are allowed to speak openly about doubts and fears, without the risk of an immediate critical reaction, go on to make better decisions than groups in which doubts have to be either hidden or dressed up as an attack on someone in particular.

Step two. Explicitly separating which trust had been damaged—in competence or in integrity—and tailoring the response to each one separately. Drawing on the Mayer, Davis and Schoorman model and on the finding by Kim, Ferrin, Cooper and Dirks that failures of competence and failures of integrity require different response rhetoric, the chairman explicitly sorted the complaints into two categories. On competence (“an eight-month delay and a twenty-two percent overrun are a failure of project management”), management gave exactly what works for this type of damage: a direct, unqualified admission of the mistake—“we underestimated the contractor risks in three key areas; here is exactly what they were; here is what we are changing in our controls.” On integrity (“you have been hiding the extent of the delay since the summer”), management presented the results of the independent review without trying to justify itself emotionally—specific dates and documents showing exactly when each figure became known to management and when it was passed on to the board—that is, it chose not an apology but a factual rebuttal, backed by an independent source rather than its own word. Mixing these two responses in November—a general, blurry defense on both fronts at once—had been one of the key reasons why the meeting had not moved the board a single step closer to trust.

Step three. Translating general anxiety into specific, measurable risk criteria. After the round of anxieties, management presented not the entire revised financial model but, first, a short list of five specific, verifiable questions that had been raised, in different wordings, by several board members: at what oil price the project pays off within a reasonable period; how likely further tightening of regulation is, according to independent legal advisers; what exactly caused the eight-month delay, and whether the cause is a one-off or will recur; what budget reserve is needed to stay within the budget in the worst-case scenario; and what will happen to the company if the project is stopped right now—that is, the cost of stopping itself, not just the cost of continuing. Daniel Kahneman, who studied decision-making under uncertainty, showed that people systematically substitute a simpler, emotionally charged question (“can we trust the people running it?”) for a hard one (“should the project continue?”)—and this is exactly the substitution that took place at the November meeting. Translating general anxiety into five narrow, verifiable questions is a way of bringing the board back from an emotionally charged question about trust to a solvable question about risk, without denying that the question of trust is also real and is already being dealt with separately, in the previous step.

Step four. Separating the decision on trust in management from the decision on the project’s fate—and taking them one at a time. Before the process was rebuilt, the two questions had been fused into one: a vote on the project was tacitly perceived by everyone as a vote of confidence in the CEO. This was a mistake, because it forced the board either to back management entirely out of politeness or to reject it entirely out of accumulated irritation—with no way to say “we are unhappy with how we were kept informed, but we agree with the plan for the project,” or the other way around. The chairman explicitly split the two votes into different agenda items: separately, an assessment of the quality of management’s reporting to the board over the past six months (with a specific decision: to introduce monthly rather than quarterly reporting on major investment projects, with a mandatory section on “deviations from plan and their causes”); and separately, a decision on the project itself based on the five risk criteria from step three. Separating these questions removed the hidden pressure and allowed the board to criticize the reporting process without turning that criticism into a vote on replacing the CEO—something the board was not in fact ready for and which would only have increased uncertainty.

Step five. Letting the board make a decision broken into stages, rather than in one go for the entire remaining life of the project. Instead of the question “continue the whole project or close the whole project,” the board was offered a decision with a three-month horizon: approve continuation of work for the next quarter under strictly defined conditions (a budget overrun of no more than five percent above the already approved new figure, mandatory monthly confirmation that the schedule was being met), with a mandatory return to a full review of the decision after three months based on actual rather than forecast data. This design rests on the same logic as the theory of behavioral self-regulation developed by the psychologists Charles Carver and Michael Scheier: when a goal is perceived as unattainable and uncontrollable, a person—and, as the practice of this meeting showed, a group too—falls into a state close to helplessness that blocks all movement. Breaking a decision that was frightening in its uncertainty into a short, verifiable step with a clear point of return gave the board back its sense of control: the board was deciding not “to the end of the project” but “the next three months, with the right to review”—and that decision proved psychologically manageable where a decision for the project’s entire horizon did not.

Step six. Marking the very first confirmed result as a public event, not as a line in a report. When, after the first control month following the December meeting, management did in fact meet the agreed conditions—the schedule deviation shrank, the budget overrun did not grow—the chairman deliberately made this fact a separate, brief item at the start of the January meeting, before moving on to new topics, and asked the CEO not merely to report the figures but to say out loud what specifically had changed in the way the project was managed. From the perspective of the neuroscience of reward, the brain’s dopamine system responds not to the fact of good news as such but to the difference between what was expected and what was received—the so-called reward prediction error. After two meetings at which the board had expected its worst fears to be confirmed, an exact match between the promised and the actual result had a stronger impact than the same result presented in a routine way would have had—and it was this that became the point from which the board’s anxiety began to subside to a manageable, working level. This is consistent with the model of organizational trust repair proposed by Nicole Gillespie and Graham Dietz: trust returns not through a one-off act but through a sequence of confirmed, verifiable episodes, and a break in this sequence at any step can easily throw the board back into its original anxiety.

The question of the independent director who had publicly accused management of concealing information at the November meeting required a separate decision. The chairman deliberately chose neither to strike the episode from the record of the meetings nor to demand an apology from the director—either option would only have reignited the conflict. Instead, the chairman met the director one-on-one and acknowledged that the anxiety behind the sharp statement had been well founded, even if its form had made the situation worse—and invited this very director to join a small working group of the board that, together with management, formulated the five risk criteria from step three. This decision rests on the same logic of psychological safety that underlies the first step: a person whose anxiety has been heard and given a channel to be applied to the task is much easier to bring back to constructive work than one whose anxiety was ignored or publicly condemned for the form in which it was expressed.

All six steps were built not into a one-off crisis meeting but into the standing rules of procedure of Meridian’s board of directors: a round of openly naming anxieties at the start of any meeting involving high uncertainty became standard practice. The procedure for preparing materials was revised separately: the finance department and the internal audit department were given a direct instruction to include in any report not only the current state of the metrics but also an explicit list of deviations from plan with an honest statement of their causes—and the chairman committed to publicly thanking management for promptly admitting a problem, not just for good results, so that no incentive remained to hold back bad news until the last moment.

The result

Below are illustrative, composite figures typical of companies of this size and profile; they do not describe any specific existing organization but show the order of magnitude achievable in similar situations over two to three quarters of systematic work on the board’s meeting process.

Two quarters after the meeting procedures were rebuilt, Meridian’s board of directors made its final decision on the refining complex without a single vote ending in a split: in contrast to the previous practice, in which decisions on major projects were often taken by a margin of one or two votes, followed by complaints behind the scenes, the key votes in January and April passed with the support of at least nine votes out of eleven. The average time between management recording a deviation from plan on a major project and the board learning about it fell from the previous two and a half to three months (under the old quarterly reporting, where unpleasant figures were also smoothed over until the next report) to three to four weeks under the new monthly format with its mandatory section on deviations. The length of meetings on contentious topics was cut almost in half—from four and a half hours at the November meeting to just over two hours on average by April—while, according to an independent assessment by the invited consultant, who conducted structured interviews with board members before and after, the quality of discussion improved: the share of board members agreeing with the statement “at meetings I have enough opportunity to honestly voice a doubt without being seen as an opponent of management” rose from twenty-seven to seventy-one percent.

Meridian’s share price had fully made up the value lost in November and December by the end of the first quarter—rising fifteen to seventeen percent from the December low, roughly in line with the movement of the rest of the sector; in other words, the company stopped losing more than the market precisely because of the board’s distrust of management. The refining complex itself was completed by mid-year within the revised, more conservative budget, with a final deviation from it of three percent instead of the original twenty-two—not because the project had suddenly become easier, but because a shorter, verifiable control cycle (“a quarter, then a review”) made it possible to spot and correct two serious schedule deviations two to three months earlier than would have happened under the old quarterly reporting format.

One point deserves a note of its own: none of the six steps eliminated the uncertainty itself—the oil price, the regulatory environment and market conditions remained as unpredictable as they had been in November. What changed was not the external environment but the board’s ability to make decisions within this uncertainty without getting stuck in anxiety and without discharging it through mutual accusations. It was this, and not an improvement in market conditions, that was the management outcome of the whole story—and that is why Meridian came out of the refining complex saga not with a temporary truce until the next piece of bad news but with permanently working rules of procedure that will withstand the next difficult quarter too.

An organization is not one person on a larger scale. It is clockwork: a multitude of gears of different sizes turning at different speeds, each running through its own cycle of Search, Decision, Action and Reward. When the market hits a company once but hard—say, a competitor uses artificial intelligence (AI—computer systems capable of solving problems that used to require human thinking) to cut its price by four-tenths—what passes through the mechanism is not one shared emotional cycle but a dozen different ones, in different workshops, departments and offices, and each gets stuck in its own phase in its own way: the design engineers in a drawn-out fear of deciding, the salespeople in anger that hits not the target but the customers, the board of directors in a stupor before the numbers. The transformation stalls not because the company has no plan but because the plan was written for a single gear, while all of them have to turn—at different speeds and in different phases. A leader’s task in such a situation is not to “motivate everyone at once” but to diagnose the mechanism part by part: to see which gear is stuck where and to intervene in each one separately, synchronizing them not at a single point but in a single direction of movement.

The situation

The company—let’s call it Versta—is a generalized, composite example of a mid-sized manufacturer: about seven hundred employees, making assemblies and components for industrial equipment—gearboxes, drive mechanisms, automation elements for conveyor lines. Its customers are large machinery makers and agribusiness companies; contracts are long, relationships have been built over years, and competition used to be moderate: in this market, price was determined mainly by the cost of metal, energy and the labor of design engineers, and custom design took weeks.

In the spring, at an industry tender, Versta loses for the first time to a competitor—a company that just three years earlier was considered noticeably weaker in engineering—with a price gap of almost forty percent. Two months later, the same competitor wins two more large contracts from Versta’s regular customers, one of them from a client Versta had worked with for almost fifteen years. An analysis of the situation shows that the cause is neither dumping nor cheaper raw materials. The competitor has rebuilt its engineering design around artificial intelligence systems that, in hours rather than weeks, generate and calculate dozens of design options for an assembly, immediately selecting those that are cheaper to manufacture while keeping the required strength and reliability. Where a team of design engineers used to need two to three weeks of work, the competitor needs two or three days of work by a single engineer who checks and approves the options the system proposes. This is not a one-off discount to break into the market—it is a new cost structure that the competitor can sustain permanently.

What happened to Versta is not an isolated, local case but part of a much broader shift that by mid-2026 is already clearly visible across the industry, especially where manufacturers compete with players from China. According to the International Federation of Robotics, in 2024 alone China installed 295 thousand industrial robots—54% of all global installations for the year—and the country’s operational stock of robots exceeded two million units and remains the largest in the world. China’s Ministry of Industry and Information Technology reports more than 30 thousand “smart factories” and a national “AI + Manufacturing” plan running through 2027. For a company like Versta, this means that a collision with a competitor armed with AI-based design systems is neither an accident nor one player’s temporary lag but a systemic, industry-wide shift that in the coming years will affect far more companies than Versta alone. But there is also a sobering caveat here, one that Versta’s board of directors did not learn right away: a study by the Massachusetts Institute of Technology (Project NANDA—Networked Agents and Decentralized AI—2025, about 300 deployments analyzed) found that 95% of companies’ generative AI pilot projects deliver no measurable impact on profit. Adopting a technology quickly does not in itself mean it will automatically pay off: what matters is not just buying the tool but rebuilding people’s real work around it, and it is precisely in this difference between “bought” and “rebuilt” that Versta’s story lies.

Versta’s CEO calls an extended meeting with the heads of all key areas—production, the design office, sales, finance, HR. The question is put with the utmost clarity: either within a year and a half to two years the company moves to a comparable model of work, or it loses market share that, at this pace, will be impossible to win back. Formally, the decision is unanimous: similar tools must be introduced and the design model changed. A budget is allocated, a person is put in charge—the deputy technical director—and the first licenses for specialized software with AI features are purchased. By every formal sign, the transformation has been launched.

Eight months later, the picture is bleak. Formally, the project is underway: the software has been bought and installed, some of the design engineers have been trained, there are even several pilot projects. But the actual share of orders designed with the new tool is under five percent. The design office keeps working the old way, citing the claim that “the tool isn’t ready yet,” although, by the deputy technical director’s own assessment, there are no longer any objective technical obstacles. The sales department does not tell customers about the move to the new design model and keeps positioning the company the old way, even though faster, cheaper design is the only argument capable of winning back some of the lost contracts. Twice in those eight months, the board of directors has put off considering a second round of funding for the program, each time on the pretext that “more data is needed.” The production director formally supports the program at meetings but informally, in conversations with shop-floor foremen, calls it “a fad that will pass.” People responsible for one and the same undertaking are pulling in different directions—and not out of malice, but each sincerely convinced that they are right.

Diagnosis through the cycle: what exactly got stuck

The first mistake Versta’s management made—and it is typical of companies whose business model is breaking down under pressure from AI-armed competitors—was the decision to talk about the transformation as if the organization had a single emotional reaction to the threat, a single pace and a single starting point. That is not so, and here it is useful to recall the very mechanism underlying productive work on a task—not as a pretty metaphor but as a sequence of quite physiological states that each person goes through individually, and along with them each department as a collection of people. First comes Search: a person or group looks around, assesses the new situation, weighs whether there are the resources to cope. This is borne out as far back as the research of the psychologist Richard Lazarus, who showed that an emotion does not arise by itself but as the result of an appraisal: does the situation matter for the person’s goals, and do they have the strength to cope with it. Then comes Decision: whether to take the matter on or not, and if so, how exactly. Next comes the Action phase—the one in which a person or team produces the result, and it is accompanied not by joy but by working tension and focus. And finally, the Reward phase: the result has been achieved and recognized, and a temporary relief sets in before the next turn of the spiral.

But this cycle has a fork right at the entrance that is often overlooked: before a person begins to search for a solution, they must pass through a primary appraisal of the threat—and here the decisive role is played by the amygdala, the brain structure described in detail by the neuroscientist Joseph LeDoux: it assesses danger quickly, before any conscious reasoning, and either passes the signal on, toward a calm search for a solution, or switches on inhibition. The psychologist Jeffrey Gray showed that moderate anxiety sharpens attention and pushes a person to look for a way out, while excessive anxiety blocks movement altogether, freezing the person in a state of wary inaction. That is exactly why the first and most common mistake management makes at the moment of an external blow is to spend effort on “motivation” and “engagement” when in fact part of the organization is physiologically still so weighed down by fear that it cannot move even to searching for a solution, let alone to action.

The same mechanism, only in different language and from a different angle, was described by the Soviet physiologist Pavel Simonov in his need-information theory of emotions: the strength of an emotional experience is proportional to how much a person lacks the information or the means to get what they need to reach a goal. Applied to Versta, this means something quite precise: the less clarity a design engineer, a salesperson or a board member has about what exactly will be required of them in the new model of work and what means they will have for it, the stronger and more destructive their emotional reaction to the very news of the change—regardless of how technically advanced the purchased tool is. A deficit not of technology but of certainty—that is what was really stuck at Versta in the first eight months, and that is exactly why buying licenses and running training sessions, for all their formal correctness, did not move this problem forward a single step.

At Versta, a working group that the CEO assembled anew—this time not to draw up yet another plan but specifically to diagnose what was stuck—determined, separately for each key department, which phase of the cycle it was actually in, rather than which phase it was supposed to be in according to the project’s overall schedule.

The design office turned out to be stuck in the fear phase at the entrance to Search. Formally, the design engineers have been trained to use the new tool. But in substance they have not even moved to the phase of searching for a solution—they are still in a state of threat appraisal, and the threat here is not abstract: the new tool essentially does a significant part of the work that an experienced design engineer used to do with their own mind and years of accumulated intuition. For a person whose professional identity is built around the ability to “feel” the right design, a tool that offers options in hours is not a relief but a direct threat to the meaning of their own expertise. The amygdala registers the danger here quite accurately—it is just that the danger is professional, not physical—and it blocks movement toward the work long before conscious reasoning about the tool’s usefulness kicks in. Talk of the tool “not being ready yet” is after-the-fact rationalization, not the real cause of the slowdown.

The sales department is stuck not in fear but in a phase physiologically described as readiness for action through irritation and anger—but aimed in the wrong direction. The psychologist Nico Frijda showed that every emotion is the organism’s readiness for a particular type of action, and in this sense anger is not destructive by nature but is fuel for overcoming an obstacle. Versta’s salespeople are angry—but angry at the design engineers (“they’re dragging their feet,” “they haven’t adapted”), at management (“they didn’t explain what to tell customers”), at the very fact that the rules of the game have changed in a market they spent years building. This irritation is a working, valuable resource: research by the psychologist Eddie Harmon-Jones shows directly that a state akin to anger or the thrill of the chase is physiologically linked not to a wish to retreat but to movement toward a goal. Versta’s problem is not that the salespeople are stuck in inaction—on the contrary, they are active, but their activity is aimed not at the new task (explaining the new, faster and cheaper model of work to customers) but at finding someone to blame inside the company. The anger is there, but it is not connected to the right goal.

The board of directors is stuck at the entrance to the Decision phase—in a state that, in terms of the theory of behavioral self-regulation developed by the psychologists Charles Carver and Michael Scheier, can be called an inability to commit to a new goal in place of the old one. Formally, the board approved the budget, but in substance it twice postponed the decision on continuing the funding—a telltale symptom that the old goal (preserving the previous pricing model and market positioning) has not yet been let go, and the new goal has not yet been fully accepted, in earnest, with a readiness to bear responsibility for the consequences. This is exactly the intermediate state that the psychology of self-regulation calls “goal disengagement”—it precedes setting a new goal, but in itself it is extremely uncomfortable and produces not paralysis in the literal sense but an endless request for “more data,” which in practice serves as a way of putting off a painful decision.

The production director is in a phase most closely described as a protracted denial of the situation’s significance—that is, returning to the psychologist Klaus Scherer’s appraisal theory, at the level of initial appraisal he has decided for himself that what is happening does not directly concern his area of responsibility (“a fad that will pass”), and so he does not even launch the phase of searching for a solution. This is the most dangerous kind of getting stuck, because it does not look like resistance—it looks like calm, and that is precisely why it goes unnoticed by management the longest.

What kept the board of directors stuck longer than one might expect from experienced, rational people deserves a word of its own. The psychologist Daniel Kahneman, in his later work on the quality of judgment (co-authored with Olivier Sibony and Cass Sunstein), showed that besides systematic biases in decisions there is also “noise”: unexplained variability in the judgments of the same people about what is essentially the same situation, depending on their mood, the order in which the items on the agenda are discussed, even the time of day of the meeting. At Versta, this noise showed up concretely: at one board meeting, where the funding question was the first item on the agenda after a healthy quarterly report, the mood was noticeably more favorable to the project than at another, where the same question was discussed last, after a difficult conversation about overdue receivables. Formally, the decision was being made on the basis of the same data about the competitive situation, but the actual readiness to make it depended more on the context of the meeting than on the substance of the question—a sure sign that the issue was not a lack of arguments but an unfinished inner process of letting go of an old, familiar goal.

Putting these four diagnoses together, the working group saw the main thing: Versta did not have a single problem of resistance to transformation. It had four different gears of the clockwork, each stuck in its own phase of the cycle, and until this diagnosis management had tried to solve all four problems with one universal key—meetings where everyone was told the same thing: “We need to change faster.” For the design engineers, stuck in fear, this sounded like an intensifying threat. For the salespeople, whose anger was not connected to a goal, it sounded like an abstract slogan with no specific task. For the board of directors, which had not let go of the old goal, it sounded like pressure that called for even greater caution. And only for the production director, who denied the significance of what was happening, did it not register at all.

The solution, step by step

Versta’s CEO abandoned a single transformation plan in favor of a differentiated intervention—a separate one for each stuck gear, but subordinated to one common direction of movement. John Kotter’s approach to change management warns that organizational change fails first and foremost not because of a poor strategy but because people’s emotional path through the change has not been taken into account—and this path differs from one group of people to another rather than being the same for all.

Step one. Naming the threat explicitly for each group, in its own words—instead of the general slogan “we need to change.” For the design office, the CEO and the deputy technical director held a separate, closed conversation with no outsiders present, in which they openly acknowledged what had been skirted until then: the new tool really does change the content of a design engineer’s profession, and this is not a cause for shame but a real shift that the entire engineering industry is going through. Acknowledging the threat out loud, from the top, by the top executive personally is a necessary condition for what the researcher Amy Edmondson calls a team’s psychological safety: until the leader has named the fear and thereby made it legitimate, an employee cannot talk about it honestly without risking looking weak or behind the times. After this conversation, the design engineers were offered not “mastering the tool” but a redefined role: not an executor replaced by a program but an expert who selects, from the options the system proposes, those that will withstand real-world operation, and who puts their own name to that decision. The meaning of the role had not disappeared—it had shifted, and this was the first thing that needed to be said out loud before demanding a move to the phase of searching for a solution.

Step two. Connecting the sales department’s anger to the right goal instead of trying to extinguish it. Instead of meetings where the salespeople were told they needed to “accept the changes,” they were invited to the design office to see the first pilot projects with their own eyes—not as observers but as participants, formulating customer requirements for the system directly. The irritation that had previously been looking for someone to blame inside the company got a specific, achievable task: to develop a new, honest case for customers explaining why the solution is now faster and cheaper with the same level of reliability, and to be the first to test this case with three of the least risky, most loyal customers. The task was framed not as “support the transformation” but as a specific, measurable challenge with a clear deadline—six weeks of preparation and three pilot negotiations. Irritation aimed at an achievable goal is, physiologically, exactly the energy that pushes toward a result faster than calm agreement does.

Step three. Giving the board of directors what its members really lacked to move from “disengagement from the old goal” to accepting a new one—not more data in general but one specific, verifiable fact. Instead of yet another presentation full of forecasts, management offered the board a specific experiment with a fixed deadline: two months, three pilot projects carried out entirely under the new model from start to finish, with an open comparison of time, cost and quality of the result against the last three projects done under the old model. This broke an overwhelming decision—“change the company’s entire model”—into a decision of a much smaller scale that the board was psychologically able to make at once: “fund a two-month experiment with a clear checkpoint.” This move is consistent with research on decision-making under uncertainty: people find it easier to make a decision when it is broken down into manageable steps with a clear checkpoint than when they are asked for a single leap of faith into an uncertain future.

Step four. Naming the production director’s position separately and directly—not as sabotage but as an unfinished phase of appraising significance. In a one-on-one conversation, the CEO did not argue with the fact that production had not yet felt the changes directly—that was objectively true: at this stage the changes concerned design first of all, not the shop floor itself. Instead of an argument, the production director was shown a calculation: if contracts kept being lost at the current rate, within a year and a half capacity utilization would drop so far that shifts would have to be cut. An abstract “fad” turned into a specific threat to his own area of responsibility, one that mattered to him personally—and it was this, not general appeals, that finally launched his own Search phase: he himself proposed preparing production, in parallel, to work with a wider range of small-batch orders, which fast design opens up.

Step five. Introducing a shared but not uniform rhythm of review—a regular “phase map” across all key areas instead of a single overall project schedule. Once every three weeks, the working group briefly recorded not “done or not done according to plan” but which phase of the cycle each area was actually in: Search, Decision, Action or Reward. This allowed management not to demand the same pace from everyone at the same time but to see where getting stuck was lasting longer than normal and called for a new targeted intervention, and where an area had already gathered speed on its own and needed not a push but resources. Such a map is a practical embodiment of the idea that a department’s emotional state can be measured and worked with systematically, just like any other management metric, rather than being left to chance or to the personal charisma of an individual leader.

All five steps were introduced not simultaneously by a single order but one after another, as the working group completed its diagnosis for each area—first the design office and the board of directors, as the stuck points most critical for further movement, then the sales department, and last of all production, since its getting stuck was the least urgent in terms of time, though also the hardest to notice.

It is important to stress what was deliberately absent from this sequence. There was no attempt to even out the pace of all four areas artificially, to fit them to a single schedule just because that looks neater in a presentation to the board. Clockwork does not work that way: the small seconds wheel and the massive hour wheel turn at different speeds not through the clockmaker’s oversight but because it is precisely the different speeds that make the whole mechanism keep accurate time. The design office went through its fear phase and reached productive Action faster than the production director managed to get through his denial phase—and that was normal, because the CEO did not demand a simultaneous finish; he demanded that every gear move in one common direction, even if at its own speed, and he did not let any of them stop for good.

The result

The figures that follow are an illustrative, composite example typical of manufacturers of this size and profile that have faced the collapse of their pricing model under pressure from AI-based competitors; they do not describe any specific existing company but show the order of magnitude achievable in a year and a half of systematic work.

Eighteen months after the differentiated work on the stuck points began, the share of orders designed using the new tool had grown from under five percent to seventy-four percent of all new projects in the design office. The average design time for a standard assembly had dropped from two or three weeks to three or four days, which allowed the company to reach a design cost comparable to the one that had let the competitor cut its price by forty percent. The share of tenders Versta won against its key competitor over the last six months of observation rose from one in seven to five in nine—that is, the company not only stopped losing market share but began to win back some of the positions it had lost earlier, including one of the three large customers that had gone over to the competitor at the start of the story.

The indicators for each of the four stuck points also changed noticeably, and in different ways, which confirms that getting the organization unstuck as a whole required separate rather than uniform work. The share of design engineers who initiated projects through the new tool on their own, without a direct instruction from a manager, rose from almost zero to fifty-seven percent of the office’s staff. The number of new pitches and customer approaches from the sales department tied specifically to the new speed and cost of design grew almost sixfold over the first six months compared with the baseline, and the conversion rate of such approaches into signed contracts was about thirty-one percent—noticeably higher than the industry average for cold proposals. From the moment of the first successful two-month experiment, the board of directors approved two subsequent rounds of funding without a single postponement, even though the second and third rounds were three times the size of the initial pilot budget. Capacity utilization, instead of the predicted decline, rose by twelve percent thanks to a new range of small-batch orders that became technically feasible precisely because of faster design.

One more outcome deserves a note of its own—it is not directly reflected in the quarterly reports, yet it determines how durable the result is: the very practice of a regular “phase map” by area has remained at Versta as a permanent management tool, applied not only to AI but to any major organizational change. The company’s management stopped assuming that the organization reacts to a blow as a single whole—and this change in management thinking, in the company’s experience, proved more valuable than any individual technology rollout, because the next blow will inevitably come from a different direction, and the clockwork will have to be synchronized all over again, but this time using an accumulated, proven scheme rather than starting from scratch.

It is also worth recording what the differentiated work on the stuck points did not do—so as not to create inflated expectations in the reader. The psychologist Barbara Fredrickson showed in her “broaden-and-build” theory that positive emotions—relief after a completed phase, pride in having covered a difficult stretch of the road—broaden a person’s repertoire of possible actions and help consolidate the experience gained. But the acute joy of success fades—and unless the practice is built into the work, it will not by itself sustain long-term motivation for further work. That is exactly why at Versta, after the first notable successes—when the share of projects done through the new tool passed the halfway mark—the working group did not wind down the regular “phase map” under the slogan “job done” but deliberately kept it as a permanent practice: the joy of an achieved result fades just as predictably as any other acute emotion, and an organization that mistakes temporary relief for final victory risks falling back into being stuck at the next external blow—just from a slightly higher starting level.

When a company gets more orders than it can fulfill, the problem seems to lie in finding people: too few applicants, high salary expectations, a labor shortage across the market. That is true, but it is not the main truth. The real bottleneck almost always turns up not at the entrance to the company but just past it: people are found, hired and put to work on sites and in offices—and then a significant share of them are lost in the first months, because a new employee circles the task for too long before starting to tackle it. This is not a question of newcomers’ character, nor of how tough the selection is. It is a stuck Search phase—that very initial state of the human activity cycle in which the brain has not yet decided whether the task is worth the effort and whether it is manageable at all—stretched over weeks where, physiologically, it should take days. A company that wants to grow faster than the labor market grows cannot hire its way out of this problem. It can do only one thing: design a new employee’s entry into the work so that the Search phase shrinks to a minimum and the phase of productive action begins as early as is at all possible without any loss of quality. What follows is the story of how this was done at a construction and development company—a composite example typical of the industry as of mid-2026.

The situation

The company—let’s call it Cascade—is a generalized example of a regional full-cycle developer: from buying land and design to construction and the sale of finished housing. At the time of these events, the company has about eleven hundred employees, of whom roughly seven hundred are frontline staff on the sites (superintendents, site foremen, quality control engineers, procurement specialists) and about four hundred are on the office side (design engineers, cost estimators, lawyers, sales and transaction support specialists). Over two years, the project portfolio has almost doubled: the company has entered two new cities, taken on several large sites and won a tender for the integrated development of an entire city block. The growth is no accident but the result of a deliberate strategy: during this period, the housing market favors those who can build quickly and predictably.

The problem is that growth in the project portfolio and growth in the number of people able to carry that portfolio are two different kinds of growth, and they have diverged. Nor is this a quirk of one company: for several years now, the construction industry as a whole has systematically been hiring more slowly than the volume of work grows. According to estimates by Associated Builders and Contractors, a U.S. trade association of construction contractors, the industry is short of three hundred fifty to four hundred fifty thousand workers every year on top of the normal pace of hiring—that is, the shortage is not a one-off but structural, reproducing itself year after year. What is more, according to joint research estimates by the Home Builders Institute (HBI) and the University of Denver, the shortage of skilled workers costs U.S. home building about ten point eight billion dollars in losses every year, because of which about nineteen thousand homes that the market was ready to buy were not built in 2024. The Russian picture by mid-2026 is milder but moving in the same direction: according to the survey of businesses that the Bank of Russia conducts regularly, the labor shortage in the country’s economy is gradually easing compared with the peaks of recent years but remains substantially above 2020–2022 levels—that is, there is some relief, but the structural problem has not gone away. In this sense, Cascade is not an exception to the rule but a typical case of an industry that is growing and, at the same time, physically cannot grow enough hands of its own to carry that growth.

Over a year and a half, Cascade hired about three hundred new employees for frontline and engineering positions. By the end of their first quarter with the company, just over half of them remained—one hundred sixty-five people; the rest either left on their own or were judged unable to cope and dismissed at the site manager’s initiative. Every position filled costs the company more than it did three years ago and takes longer—competition for engineering talent on the labor market is intensifying also because manufacturing, which is going through its own wave of technological re-equipment, is fighting harder and harder for the same people.

The HR director brings the board of directors a figure that changes the tone of the conversation: the average cost of a single engineering or technical hire—including job advertising, recruiters’ work, paperwork, training and the lost productivity of a site that temporarily operates without the person it needs—has almost doubled in two years. But even the rising cost of hiring was not the main cause for alarm. The main cause was that, with early attrition at this level, the company was effectively hiring the same person two or three times for the same position, paying for it anew each time—while the project portfolio was growing faster than the company could plug this leak.

Management’s first reaction was quite predictable for the labor market of the time: raise salary offers above the regional average, widen the recruiting funnel, bring in staffing agencies and start targeted recruiting at specialized colleges and universities. All these measures were right and far from useless—the funnel did widen, and there were more applicants. But the share of those who left within the first three months barely changed. The company got faster at finding people and kept losing them just as fast. It became clear once and for all: the bottleneck was not at the entrance to the company but just past it.

Tellingly, a similar picture was taking shape on the office side too, although turnover in the first months was lower there than on the sites—about a third versus almost half among frontline staff. The difference in scale was noticeable, but the nature of the problem turned out to be the same: a young cost estimator or design engineer who had come from another company or straight out of school spent the first weeks not producing estimates or drawings but figuring out in which of three parallel document templates this particular organization keeps its calculations, which of the managers actually approves decisions and who merely signs off on them formally, and whether it is even all right to ask questions without looking incompetent next to colleagues who have worked there for ten years. The HR director later admitted that it was precisely the match between this picture in two such different parts of the company—on the construction site and in the design office—that became the final argument that the issue was not the specifics of a profession but the overall design of the very entry into the organization.

Diagnosis through the cycle: what exactly got stuck

To understand what happens to a new employee in the first weeks on the job, it helps to break their path down into the same states that any person goes through when taking on an unfamiliar task—this is not a metaphor but a sequence described quite precisely in science. First comes the Search phase: the person looks around and assesses what is expected of them, whether they have the resources to cope, and whom to turn to if something is unclear. This is not idle curiosity but necessary work of the psyche: as the psychologist Richard Lazarus showed long ago in his cognitive appraisal theory of emotion, before a person starts to act, they appraise the situation—is it significant, is it manageable, what could a mistake cost. Until this appraisal is complete, a person is physiologically not ready for confident action, however qualified they may be on paper. Then comes the Decision phase: the person in effect says to themselves, “Yes, I’ll take it on,” and this inner consent also takes time and a degree of certainty about the situation. Next comes the Action phase, in which the result is born, and only after it the Reward phase, when the result has been noticed and acknowledged and the person can move on with a clear conscience to the next, slightly harder task.

For an experienced employee who has long worked at the company, this entire sequence takes seconds or minutes: they already know the context, they don’t need to reassess whom to trust and whom not to, which rules are written and which are not. In a new person, the Search phase is not compressed—it is unfolded to its full width, because absolutely everything is unknown at once: they don’t know the people, don’t know the unwritten rules of the site, don’t understand who actually makes the decisions and who simply talks loudly at meetings, and are not sure whether they are allowed to take the initiative or must first ask permission for every step. And this is where what the neuroscientist Joseph LeDoux described about how fear works comes into play: the amygdala—a brain structure that assesses potential threat quickly, even before slow, conscious reasoning kicks in—reacts to uncertainty in exactly the same way as to direct danger. A new employee on an unfamiliar site, where it is unclear who is right in a disputed situation and what will happen if they make a mistake, is physiologically in a state of heightened anxiety almost constantly, even if outwardly they look calm and nod along at the safety briefing. And the behavioral inhibition researcher Jeffrey Gray showed that moderate anxiety sharpens attention and helps avoid gross mistakes, but anxiety above a certain threshold simply blocks action—the person hesitates, asks again instead of doing, and puts off what they could have done on their own.

When Cascade’s working group—the HR director, the head of occupational health and safety, three experienced site managers and a training specialist—walked through the cyclical “Search—Decision—Action—Reward” logic for each of the employees who had recently quit or been let go, the picture came together quickly, and it differed markedly from the original explanation of “wrong person, wrong qualifications.”

First, it turned out that a new Cascade employee’s Search phase physically stretched over four to six weeks—because training and immersion in the work were set up as a sequence of separate, loosely connected events: an introductory safety briefing on the first day; familiarization with the project documentation—whenever the engineer who was supposed to walk them through it had some free time; getting to know the people on site—ad hoc, as things came up; access to the necessary software and approval systems—sometimes only after weeks, because the access request went through a chain of several signatures. As a result, a new site foreman formally started work on day one but actually began deciding anything on their own only in the third or fourth week. All this time they were on the books as working and drew a salary, but the productive Action phase had not yet begun for them—they were still in the Search phase; it was just that no one named this state out loud or deliberately shortened it.

Second, a gap between Decision and Action came to light, caused by the absence of clear permission to act independently. New engineers and foremen had no clear sense of the boundaries: what they were entitled to decide on their own and what they had to clear with the site manager. In the absence of that clarity, many chose the safest path—clearing everything, even the obvious things, out of fear of making a mistake in a new place. This sharply slowed their own work and, more damagingly, irritated the experienced site managers, who were already short of time: they saw not a proactive specialist but someone who “asks again about every little thing,” and drew the hasty conclusion that the newcomer was poorly qualified—although in fact the problem lay not in qualifications but in the absence of clearly defined authority.

Third—and this gap came to light only after direct conversations with several workers who had quit in their first months—a problem surfaced with the Reward phase, or more precisely with its almost complete absence in the first weeks. A new employee, even one who had done something useful and right in their first days, very rarely received clear feedback that it had been done well. Feedback on Cascade’s sites worked by exception: silence meant “everything’s fine,” and a conversation with the manager almost always began only when something had gone wrong. The psychologist Nico Frijda, who studied emotions as the organism’s readiness for a certain type of action, showed that without a signal confirming that an action was right, a person does not consolidate the behavior pattern and goes on acting tentatively, sparingly, looking over their shoulder—that is, gets stuck in a state on the border between Decision and Action instead of moving on to confident, productive work. Put simply, if a newcomer is praised only by the absence of criticism, they have nothing to rest on to believe they are moving in the right direction, and they keep behaving like someone who is still looking around rather than someone who is already acting.

Putting all three gaps together, Cascade’s working group formulated a diagnosis directly opposite to the one they had started from: the issue was not a shortage of qualified candidates on the labor market—although that shortage is real and is confirmed by industry statistics far beyond a single company. The issue was that the path from “hired” to “a productive member of the team” took the company six to eight weeks on average, and all that time remained undesigned, left to chance—entry into the company had not been designed as a sequence that compresses the Search phase; it was set up as a random set of procedures, each of which seemed reasonable on its own, but together they stretched uncertainty out to more than a month. It was precisely in these weeks, while a person did not yet feel confident and had not yet been recognized for their first successes, that most of the early departures happened—not because the person couldn’t cope, but because they didn’t have time to realize that they were coping. The HR director summed it up for the board of directors briefly: “We don’t hire more slowly than our competitors. We lose what we have already hired faster than they do.”

The solution, step by step

Cascade’s working group made a decision of principle: not to widen the recruiting funnel any further (that had already been done and had not helped) but to redesign a new employee’s first eight weeks as a managed, measurable process of compressing the Search phase. The goal was formulated as concretely as possible: to cut the time to the first independent task acknowledged by a manager from six to eight weeks down to two. As a separate, supporting line of work, they decided to look at the technical side of the question as well: how much of the work that had previously required new, still inexperienced people could be taken off people’s hands altogether through automation—fortunately, by mid-2026 a body of verifiable evidence on this question had already accumulated. A peer-reviewed study of automation in modular home building (Ouda and Haggag, 2025) showed that automating standard, repetitive operations cuts production time per unit by almost forty percent and labor costs per unit by almost seventy percent, with a payback period of about three years. For Cascade, this finding meant not replacing people but an important adjustment to the strategy: part of the load that had led to newcomers being thrown at complex, intuition-heavy tasks too early could be removed by automating standard calculations and document workflows, freeing up the time of the mentors and of the newcomers themselves for exactly what cannot be automated—judgment, responsibility, live communication with the crew.

Step one. Splitting positions into streams and building, for each stream, a map of the first eight weeks by day rather than by month. Instead of a general induction course, the same for everyone, they created three separate tracks—for site foremen, for quality control engineers and for estimating engineers—because what a superintendent on site needs to know in the first week and what an estimator in the office needs to know overlap by barely a third. For each track, they spelled out day by day what the person should learn, whom to meet, which system to get access to, and which first small task to complete independently and exactly when. This change, technical at first glance, turned out to be the most important one: it turned the first weeks from a vague “find your feet” into a clear sequence of steps, and a clear sequence is exactly what shortens the Search phase, because it removes the very uncertainty to which the amygdala reacts so sharply.

Step two. Assigning every new employee a mentor from among the experienced staff—with an explicit separation between the roles of mentor and direct manager. Before this, the site manager had to set tasks, answer all of the newcomer’s questions and keep track of the site plan all at once—and the newcomer’s questions inevitably lost out in priority. Now every new employee had a mentor—an experienced colleague in the same line of work but not their direct boss—whom they could approach with any question without feeling that it was a distraction from “real” work or made them look incompetent in front of the person evaluating their work. Mentoring was paid with a separate allowance, albeit a small one, and counted in the mentor’s own performance review—this mattered so that the role would be seen not as an extra burden without recognition but as part of the job that was also noticed and rewarded.

Step three. Introducing early, specially designed small tasks with a guaranteed result in the very first days. Instead of spending weeks “sizing up” the work, a new site foreman received, as early as the second or third day, a small, clearly defined task entirely within their authority—for example, independently conducting and documenting one of the scheduled site walk-throughs against a checklist—with an unambiguous, visible result. This decision rested on an understanding that the psychologist Nico Frijda wrote about in connection with emotions as readiness for action: a person needs not an abstract explanation that they are capable of doing the work but a concrete, if small, completed cycle—from receiving the task to finishing it—that physically proves: I can. Each such early small win went through the cycle in full in a single day rather than hanging in limbo for weeks, and it was precisely this full, fast cycle, repeated many times over in the first days, that made up the compression of the Search phase.

Step four. Explicitly spelling out the boundaries of independent decisions for each position—what can be decided without clearing it and what requires a conversation with the manager. For site foremen, quality control engineers and estimators, they drew up short lists that literally fit on a single page: which decisions within their area of responsibility an employee makes on their own, without asking, and which they must clear. This eliminated that very defensive strategy of “clear everything for fear of making a mistake,” which had previously disguised a newcomer’s insecurity as excessive caution. As soon as the boundaries became clear, appraising the situation—the very appraisal Richard Lazarus wrote about: is it significant, is it manageable, what could a mistake cost—began to take not hours of doubt but seconds, because the answer to the question “am I entitled to decide this myself?” no longer required guesswork.

Step five. Introducing mandatory, immediate feedback on the first weeks—instead of a silent default of “everything’s fine.” Site managers and mentors were required, at the end of each day of a new employee’s first two weeks—and then once a week until the end of the second month—to talk not only about what needed fixing but first and foremost about what had been done right, and why exactly it was right. The format was deliberately kept short, three to five minutes, so as not to turn it into an extra bureaucratic burden, but mandatory, so that it would not get lost in the daily grind. This directly closed the loop on that very Reward phase whose absence had previously kept new employees in a state of tentative, sparing behavior much longer than was objectively necessary.

Step six. Automating some of the standard operations that used to eat up the time of both newcomers and their mentors—not to cut headcount but to free up time for real engagement in the work. Drawing on validated calculations for automating standard production and calculation operations, Cascade switched part of its repetitive cost-estimate calculations and the filling-in of standard quality control documentation to automatic mode—work that a new estimator or engineer used to do by hand for weeks while learning both the profession itself and a dozen internal templates at the same time. This did not replace the new employee; it removed the dullest and most impersonal part of the work from their first weeks, leaving time for what cannot be automated—a conversation with the crew on site, an independent walk-through, hands-on mentoring.

Step seven. Making turnover in the first three months a separate metric visible to management—rather than part of the overall HR statistics. Before this, the share of early departures was lost in the overall annual turnover figure and did not draw the separate attention of the board of directors. Now the metric for reaching independent work and for retention in the first three months began to be reported monthly, as a separate line, on a par with construction deadlines and the projects’ financial indicators—that is, it gained the same managerial weight as the production metrics that a construction company traditionally considers the “real” numbers.

All seven steps were introduced one after another over the course of a single quarter, starting with the cheapest and quickest—clarifying the boundaries of authority and introducing daily feedback—and ending with the more labor-intensive changes: day-by-day maps for each stream of positions, automation of standard calculations, and a mentoring system with separate pay, which required budget approval.

The result

Below are illustrative, composite figures typical of a regional construction and development company of this size; they do not describe any specific existing organization but show the order of magnitude achievable in two or three quarters of systematic work on a new employee’s first weeks. The real industry statistics mentioned above (the construction labor shortage in the United States, the impact of automation in modular production, the trend in the labor shortage according to Bank of Russia surveys) are verifiable data given for industry context; they do not relate to Cascade directly.

Nine months after the process overhaul began, the average time from starting work to the first task completed independently and acknowledged by a manager fell from six to eight weeks to twelve to fourteen days. The share of new employees quitting or being let go in their first three months dropped from forty-five percent (one hundred thirty-five of the three hundred hired over the previous year and a half) to roughly eighteen to twenty percent—that is, to a level comparable with the industry average for companies with a well-run onboarding process. The average cost of one successful hire who stayed—given that the company now has to rehire for the same position much less often—fell by roughly a third compared with the period when the company simply widened the recruiting funnel without changing the onboarding process.

They also measured something they had not originally planned to measure: how quickly a new construction site reached its planned output. Where at least half of the renewed team (the foreman, the engineers, some of the frontline staff) was working on site, the planned deadlines for interim stages began to be met noticeably more accurately—schedule slippage caused by a shortage of people on site ready to work independently fell by roughly half. In other words, the impact of compressing the Search phase did not stay inside the HR department but was directly reflected in the company’s main production indicator—construction deadlines, for which the whole organization is ultimately responsible.

The automation of standard calculations and document workflows, introduced in parallel with the overhaul of the onboarding process, had a smaller impact, but one that matched the industry data in direction: the time a new estimator spent preparing a standard estimate fell by about a third as early as the end of the second month on the job—not because the person had gained experience faster, but because part of the mechanical work simply stopped being their job, and from the first weeks they were working on what really required their judgment.

The cost of the program itself—the onboarding maps for each stream, the mentors’ allowance, managers’ time for weekly feedback, setting up the automation of standard calculations—turned out to be noticeably lower than the cost of the expansion of the recruiting funnel alone that had originally been discussed as the main measure. In other words, the company solved the problem not by spending more on attracting people but by losing fewer of those it had already attracted—and keeping a person you have hired is almost always cheaper than finding a new one.

An important caveat, typical of stories like this one: some new employees kept leaving even after the process was overhauled—some because a construction site, objectively, is not the pace or the working conditions that suit them, others for personal reasons unrelated to work. The point of compressing the Search phase is not to retain every single hire without exception but to make sure that a person decides whether to stay or go consciously, having already tried themselves in real work and received honest feedback on their results—rather than leaving after three weeks without ever really understanding whether they could make it in the position or not. And Cascade’s working group itself acknowledged one more thing worth stressing: compressing the Search phase does not eliminate the structural labor shortage in the industry—it does not create new people on the labor market where they physically do not exist. It solves a different but no less important problem: it stops throwing back onto that same scarce market the people the company has already found and already paid for.

The first three days after a serious incident decide more than the strategy for the next three years—because it is in these seventy-two hours that an organization either moves through the emotional cycle in concert, as a single organism, or breaks apart into isolated pockets of panic, each of which starts its own unsynchronized cycle of fear. Managing a crisis in these hours does not mean suppressing the emotions of employees, clients and the board of directors. It means setting up a command center that takes the organization through the phases of the cycle—Search, Decision, Action, Reward—on a clear schedule, rather than leaving that passage to chance. Whoever builds a structure in the first twenty-four hours that keeps panic from spreading out of control, yet does not hold back the normal human reaction to a shock, comes out of the crisis stronger. Whoever tries to manage a crisis through the top executive’s heroic manual intervention, with no structure, loses not only their reputation but also the people who could have saved it. What follows is the story of an aviation and logistics operator whose infrastructure and coordination failed at the same time in a single night, and of the command center protocol that turned seventy-two hours of managed chaos into seventy-two hours of a managed cycle.

The situation: the night when not just one thing failed, but everything at once

Picture a mid-sized aviation and logistics operator—let’s call it Aurora Cargo, because this is a composite, illustrative case built from the typical features of companies of this kind, not a specific existing organization. A fleet of cargo and passenger aircraft serving regional and mainline routes, its own logistics hub at one of the major hub airports, contracts to carry perishable goods and medical products, and annual revenue in the range of ₽40–55 billion. For many years the company had taken pride in on-time performance above the industry average and in its reputation as a reliable carrier with which one could build long-term logistics chains.

The crisis began not with one event but with three overlapping ones. Around midnight, one of the aircraft on a route with a partial passenger load had a serious technical incident on approach—the crew coped, no passengers or cargo were harmed, but the aircraft made an emergency landing at an airport that was not one of the company’s bases, and flights were fully halted for several hours for a technical investigation. At the same time, within the same hour, the information system for cargo allocation and flight connections failed at Aurora Cargo’s main logistics hub—technically unrelated to the incident in the air, but coinciding with it in time and practically paralyzing the hub’s dispatch operations. And the third overlap, the most destructive in its impact: passengers on board wrote about the incident on social media almost instantly, even before the company’s official statement—with exaggerated and in places inaccurate details, including the phrase “we nearly crashed,” which gathered hundreds of thousands of views within just two hours.

By four in the morning, Aurora Cargo’s CEO—let’s call him Dmitri Andreevich—had on his desk (figuratively speaking: at that moment he was at home, woken by a call from the duty dispatcher) not one problem but three parallel ones, each moving at its own speed: a technical one (investigating the causes of the incident and the safety of continuing to fly similar aircraft), an operational one (a paralyzed logistics hub, dozens of flights and hundreds of tons of cargo without up-to-date dispatching) and a reputational one (a rapidly spreading, in places unreliable version of events in the public sphere, on which the company had not yet made a single official comment). Taken separately, any of these three situations was manageable for a company of this size—Aurora Cargo had separate procedures for each scenario. The problem was not a lack of procedures but the fact that all three were triggered at once, and three parallel response groups—the flight safety department, the logistics hub’s duty shift and the small public relations department—began acting not as a single command center but as three isolated pockets, each with its own pace and, as turned out to matter more, its own unsynchronized emotional state.

The first six hours of the crisis—from midnight to early morning—became a telling example for Aurora Cargo of how the absence of a crisis response structure multiplies the damage from the incident itself. The flight safety department, following internal procedures, focused exclusively on the technical investigation and deliberately did not get in touch with either logistics or the press office, guided by the principle of “don’t make noise until there are confirmed facts”—a principle that is reasonable in a narrow technical sense but catastrophic when the public sphere is already filling up with versions of events without the company’s input. The logistics hub’s duty shift, faced with the system failure, spent the first three hours trying to restore the software on its own, without escalating the problem above the technical support level—because the formal escalation procedure required a written report on the nature of the failure, and the nature of the failure had not yet been established. The public relations department, which at the time consisted of three people, learned the scale of the public reaction not from flight safety and not from logistics but from its own social media monitoring—about two hours later than it should have—and was forced, for the first time, to contact the CEO directly just to understand what had actually happened technically, in the fourth hour of the crisis, when the first wave of posts had already passed its peak.

By nine in the morning, when the business media and not just social networks had started talking about the incident, Aurora Cargo still had no single, agreed account of events, no single list of priority actions and no single person responsible for decisions that went beyond the authority of an individual department. Three parallel meetings—in flight safety, in logistics and in the press office—were going on at the same time in different meeting rooms of the same building, and participants learned about one another’s decisions not in real time but from later retellings. The company’s CFO, who that morning was getting calls from worried representatives of its three largest shipper clients demanding a clear explanation, could not give an answer, because he himself had not received a single, coherent picture from any of the three groups.

Diagnosis through the cycle: what exactly got stuck

If we look at the first hours of Aurora Cargo’s crisis through the emotional cycle examined in Part I of the book—Search, Decision, Action, Reward—the diagnosis turns out to be precise and, in that sense, encouraging: it was not the people or their competence that broke. What broke was the coordination among three parallel cycles, each of which was moving honestly, and reasonably in its own way, through its Search phase without knowing which phase the other two were in.

The organizational psychologist Karl Weick, who studied how people behave in acute crises, described the leader’s key role in such hours in a single word—“sensemaking”: the top executive’s task is not to personally get to the bottom of the technical details of the incident but to assemble, as quickly as possible, scattered facts arriving at different speeds into a single picture the whole organization can understand, and to pass that meaning on down the chain, rather than act alone, relying on their own understanding of the situation. At Aurora Cargo, no such sensemaking happened at all during the first six hours—not because there were no smart people ready to do it, but because there was no structure that would physically bring three parallel streams of information together at one point. Each of the three groups was honestly going through the Search phase on its own—assessing the scale of its part of the problem—but the three Search phases were out of sync, and no one in the organization held the full picture for longer than a few minutes before it changed again.

It is worth recalling here the mechanics laid out in detail in Part I of the book. The psychologist Richard Lazarus showed that an emotion arises from an appraisal—whether a situation is significant and whether there are enough resources to cope with it. Until that appraisal is complete, a person remains in a state of heightened anxious readiness rather than moving on to calm, purposeful action. For Aurora Cargo’s three response groups, the appraisal of the situation could not be completed in principle, because each group had only a third of the picture: flight safety did not know the scale of the public reaction, the press office did not know the technical details of the incident, and logistics knew neither. An incomplete appraisal of a situation is not an abstraction but a concrete physiological state: the neuroscientist Joseph LeDoux showed that the amygdala detects threat even before conscious reflection kicks in, and later research showed that its response grows stronger when the situation is unpredictable. The three isolated response pockets at Aurora Cargo were in a state of maximum unpredictability not because the threat was objectively large, but because no one could see the situation as a whole—and this sense of fragmentation in itself heightened anxiety more than the technical incident, the system failure or the unreliable posts did, each taken separately.

The specific management trap that Aurora Cargo’s flight safety department fell into—refusing to get in touch with other units until it had confirmed facts—is telling. The psychologist Jeffrey Gray described the physiological mechanism behind such a reaction: at signs of risk, the behavioral inhibition system switches on and halts movement—a useful function when it stops you from taking a rash step, but a destructive one when it is applied not to a technical decision (whether to announce the cause of the incident, which has not yet been finally established) but to the very fact of communication (whether to report at all that an investigation is under way). The safety department confused two different questions—“what exactly happened” and “should we even say that we are finding out”—and froze both at once, although the second question did not require waiting for the results of the investigation at all.

A similar but mirror-image trap held the logistics hub: the duty shift, following the formal escalation procedure, which required a written description of the nature of the failure before a higher level of support could be brought in, in effect applied a peacetime bureaucratic procedure to a situation where there was no longer time for it. This is a classic illustration of the lesson behind the creation of the Incident Command System—a management structure developed in the United States in the 1970s after California’s catastrophic 1970 wildfires, when, as later analysis showed, the response failed not for lack of people or equipment but for lack of coordination among independently operating services. The key principle of this structure—unity of command in a crisis and a span of control of about five direct reports per leader—grew precisely out of the observation that in the first hours of a serious incident, the ordinary, function-based organizational structure of peacetime stops coping not because employees are incompetent, but because it is not designed for the speed and density of decisions that a crisis demands.

What was happening to Dmitri Andreevich’s own emotional state in these first hours deserves a note of its own—because it was this state, transmitted further through the organization without conscious filtering, that set the tone for the entire crisis. The first calls from the duty dispatcher, then from the CFO, then from worried clients caught him without a single picture and forced him to give partial, uncertain answers to each group separately—a state in which the very anxiety of a top executive without a coherent picture was transmitted down the chain as the signal “even the boss doesn’t understand what’s going on,” amplifying anxiety at every level of the company rather than reducing it. The psychologist Klaus Scherer, the author of appraisal theory, showed that the appraisal of novelty and unpredictability is triggered anew by every new fragment of information that does not fit with the previous ones. In the first six hours of the crisis, Dmitri Andreevich was not going through one cycle of appraising the situation; in effect, he was starting it over with every call—because each call brought information that was not built into the overall picture and was therefore perceived afresh as a new threat.

Put these mechanisms together and the diagnosis becomes precise: Aurora Cargo was not suffering from a shortage of competent people, resources or industry procedures written in advance for each individual problem. The company was suffering from the absence of a single structure that would turn three parallel, unsynchronized Search cycles into one common cycle for the organization—with one person responsible for the whole picture, one decision-making schedule and a clear understanding at every level of which phase of the cycle the company was in right now.

The solution: a command center as a protocol for moving through the cycle hour by hour

The turning point came on the morning of the first day, when Aurora Cargo’s director of operational safety—not the company’s top executive but a middle manager who had previously done an internship in an organization built on emergency command principles—proposed that Dmitri Andreevich immediately stop the three parallel meetings and set up in their place a single crisis command center modeled on the Incident Command System he knew. It was this structure, and not the CEO’s heroic personal intervention, that became the basis of the protocol that turned the company’s passage through the emotional cycle during the crisis into a managed schedule, marked out by the hour.

Hour 0–2. One command center, one leader, one physical (or video-linked) gathering place. Hours here and below are counted from the moment the command center was assembled. The first action was not an investigation and not communication but structure: the three separate response groups were physically brought together in one room, and one person—in this case Dmitri Andreevich himself, with operational management delegated directly to the director of operational safety as the one more competent in command center structure—was appointed the single point where all three streams of information converged and from which all decisions beyond the authority of an individual department came. This is a direct application of the central principle of the Incident Command System: unity of command in a crisis, not because everyone else is incompetent, but because coordination among several equal decision-making centers physically cannot keep up with the speed of a crisis. The span of control was set up on the same principle: each head of a line of work—technical investigation, operational recovery, external communications—was responsible for no more than five or six direct reports, so as not to lose control of what was happening in their area.

Hour 2–6. Splitting the Search phase into three parallel but synchronized streams with a fixed review cadence. Instead of waiting for each department to put together an exhaustive picture of its own area, the command center set a strict rule: a short review every thirty minutes, at which each line of work reports not final conclusions but its current state of knowledge, including explicitly flagged gaps. This decision draws directly on an idea laid out in detail in Part I of the book: fear of incomplete data paralyzes not because the data are objectively insufficient, but because the incompleteness has not been acknowledged out loud as a temporary, expected state. As soon as each department was allowed to report “we don’t know X yet, but we definitely know Y” instead of waiting for complete clarity before reporting, the speed of information exchange inside the command center increased several times over, and the anxiety caused precisely by the sense of a fragmented picture began to subside—because the fragmentation became an explicit part of the common picture, shared by everyone, rather than a hidden individual burden for each separate department.

Hour 4 (in parallel with the technical investigation). The first public statement—not with exhaustive facts, but with the fact that the company knows about the situation and is managing it. Here the protocol parted ways with the flight safety department’s intuitive but mistaken decision to wait for complete clarity before any communication. The command center made a decision with the opposite logic: the first public statement came out four hours after the command center was assembled, and it contained no final conclusions about the cause of the technical incident—such conclusions objectively did not yet exist—but it did contain three specific things: confirmation that the incident had happened, without exaggeration or downplaying; information that passengers and cargo were unharmed and that an official investigation was under way; and a direct commitment to give the next update within a specific, stated number of hours. This decision rests on the same logic described by the classic trust model of Mayer, Davis and Schoorman: trust rests on three perceived pillars—competence, benevolence and integrity—and in a safety-related crisis, the company’s silence is perceived by the audience not as caution but as a blow precisely to the pillar of integrity, because silence, in the absence of an alternative source of information, is instantly filled with other people’s, often distorted versions of events. A telling counterexample from another industry is the crisis of trust in nuclear power in Japan after the accident at the Fukushima nuclear power plant in 2011, where the share of Japanese who saw nuclear power as necessary fell from 35.4% to 17.9% over seven years, while the share who distrusted it tripled, largely because of a failure of crisis communication specifically in the first days, and not only because of the scale of the accident itself.

Hour 6–24. Separating the Decision phase from the Action phase for each of the three lines of work—with an explicit criterion for the transition. The command center set a rule that removed the main trap of the first six hours: the technical investigation into the causes of the incident continues at its own pace and cannot be artificially accelerated—it is a Search phase that will end when it ends, not on the command center’s schedule. But the operational recovery of the logistics hub and external communications do not have to wait for the technical investigation to finish, because decisions in these two areas do not depend on its outcome. Logistics received its own, separate decision to move from Search to Action: within six hours, a backup manual dispatch process was set up for critical cargo (medical products and perishable goods were given priority) while the main information system was being restored by technical support—a decision made with a knowingly incomplete understanding of the cause of the failure, but enough to stop the damage from piling up. External communications received their own plan, independent of the results of the technical investigation: a series of three scheduled updates during the first twenty-four hours, an interview with a company representative for the business media at the twelfth hour, and direct, personal contact with the management of the three largest shipper clients within the command center’s first eight hours—before they had time to raise claims.

Hour 24–48. An explicit, announced transition to the Action phase across the whole organization and the first checkpoint for an objective review of results. By the end of the first day, the command center carried out what, in terms of the cycle, can be called a formal crossing of the line: it was announced—inside the company and, in adapted form, externally—that the acute phase of the crisis was over and that the company was moving from emergency response to planned recovery in all three areas at once, with specific deadlines for each. This step matters not so much for its content as psychologically: without an explicitly announced point of transition, an organization tends to stay stuck in emergency mobilization mode far longer than the real situation requires—people keep working in a state of heightened anxious readiness simply because no one has explicitly said that the acute phase is over. Richard Lazarus showed in his appraisal theory that reducing emotional tension requires not so much the disappearance of the threat as a reappraisal of the situation as manageable—and such a reappraisal is triggered most effectively by an explicit, authoritative signal, rather than by a gradual fading of anxiety that no one announces.

Hour 48–72. The Reward phase as a management action, not a spontaneous one—a debrief of what happened and explicit recognition of those who acted in the first hours. On the third day, the command center held a structured debrief of everything that had happened over the seventy-two hours—not as a search for the guilty but as an analysis of what had worked in the command center structure and what had not, including an honest admission: the system under which three departments acted in isolation for the first six hours was a real management failure, not the result of particular people’s incompetence. Separately, the duty dispatcher who raised the alarm in the first minutes and the logistics team that set up backup manual dispatching on a tight timeline, with an incomplete understanding of the cause of the failure, were explicitly and publicly recognized within the company. This step is neither a formality nor a corporate courtesy. The psychologist Barbara Fredrickson showed that positive emotions after a tense episode do not in themselves create long-term motivation, but they broaden the repertoire of learned experience and help consolidate what the organization has learned—if that experience is explicitly talked through and acknowledged, rather than simply lived through in silence and pushed aside by the next urgent task.

The order of these steps matters as much as their content. A single command center with one responsible person eliminated the very cause of the three parallel cycles falling out of sync. Permission to report an incomplete picture every thirty minutes relieved the anxiety generated not by the crisis as such but by the sense of fragmented knowledge about it. An early public statement that did not wait for complete clarity protected the integrity pillar of the trust model before silence had time to destroy it. Separating the Decision and Action phases along lines of work that did not depend on each other allowed logistics and communications to move without waiting for the pace of the technical investigation. An explicitly announced transition from emergency mode to planned recovery cut short a phase of anxious mobilization that would otherwise have dragged on of its own accord. And a structured Reward phase on the third day consolidated the completed cycle as experience rather than as a trauma the organization simply lived through in silence.

The result

All the figures below are an illustrative, composite example for an aviation and logistics operator of comparable size, not the metrics of a specific existing company.

By the end of the first day, Aurora Cargo had issued three public updates on a schedule announced in advance, instead of the single belated statement that would have come out under the old, uncoordinated approach of three isolated departments. The backup manual dispatching of critical cargo, set up in the command center’s sixth hour, made it possible to keep on-time delivery for 92% of the highest-priority cargo (medical products and perishable goods) while the main information system was almost completely down for fifteen hours. Direct contact with the three largest clients within the command center’s first eight hours, before they had time to raise claims on their own, meant that none of the three contracts was terminated and no penalty proceedings were initiated for breach of the agreements’ terms—even though, according to the internal assessment of the company’s legal department, formal grounds for such claims did exist.

The public reaction on social media and in the business media, which grew rapidly in the first four hours on the strength of unreliable and exaggerated versions of events, stabilized and began to decline within twelve hours of the company’s first official statement—according to internal mention monitoring, the share of posts with an explicitly negative, panicked tone fell from a peak of about 74% in the first six hours to 31% by the end of the second day, even though the overall number of mentions stayed high for several more days (the company remained a topic of discussion, but the discussion shifted from guesses and rumors to the company’s recovery actions).

Even more telling than the direct operational results was what happened to Aurora Cargo’s response time itself in subsequent, smaller incidents. Within a year of this crisis, the company had codified the command center protocol as a mandatory procedure for any situation affecting flight safety, operational resilience and public reputation at the same time—with clear criteria under which the protocol is activated automatically, without waiting for the CEO to decide personally that the situation is “serious enough.” Over that year, the protocol was activated four times on smaller occasions—two technical incidents with no consequences for passengers, one failure by a ground handling provider at one of the regional airports and one episode of unreliable media reports with no real technical incident behind it. The average time from the start of a situation to the first structured public statement fell from 13–14 hours in the original crisis to 65–90 minutes in the four subsequent cases—not because decisions were being made more hastily, but because the command center structure and the roles written down in advance no longer had to be built from scratch each time.

The director of operational safety who had proposed the command center model in the first hours of the original crisis later described the main shift this way: “Before this case we had a shared feeling that a crisis is when everyone tries very hard at the same time. Afterward it became clear that a crisis is when the organization needs not more effort but a different schedule. People on that first night tried no less hard than a year later, in the fourth incident. The difference was not in the effort but in the fact that the effort was finally not being wasted on a lack of coordination among ourselves.” This is an accurate description of what the command center protocol provides as a way of moving through an organization’s emotional cycle in a crisis: not the abolition of anxiety, fear or confusion in the first hours—they are natural and inevitable in a serious incident—but a structure that keeps these states from multiplying through a lack of synchronization between departments and turns seventy-two hours of uncertainty into seventy-two hours of managed movement toward recovery, marked out by phase.

The year after a crisis is the most dangerous moment in the life of a company that has survived it, because this is exactly when an organization most often makes one and the same mistake: it considers the cycle completed once and for all and goes back to its old way of working, as if nothing had happened. This is a mistake not of memory but of design. The cycle of emotional self-regulation—the sequence “Search—Decision—Action—Reward” that an individual person and an organization as a whole both go through—does not end when the crisis has passed and the numbers are back in the black. It is either built into the ongoing practice of management or starts over from scratch at the next blow, because everything the company learned in the acute phase is quietly forgotten as the tension ebbs away. The difference between a company that simply dusted itself off after hitting bottom and a company that came out of the crisis stronger is the difference between a one-off measure and institutionalization: between “we don’t do that anymore” as an emotional vow made in the heat of the moment and “we now always do it this way” as a built-in, mandatory part of corporate governance. What follows is the story of how this was done in a diversified holding company—a composite example typical of companies of this kind.

The situation

The company—let’s call it Atlas—is a generalized example of a diversified holding company with assets in industrial manufacturing, logistics and commercial real estate. Three relatively independent but managerially linked business lines, a single management company, about forty-five hundred employees across the group as a whole. Almost two years before this story begins, the holding company went through an acute crisis that people inside the company later came to call simply “the bottom”: a sharp drop in demand in the industrial business coincided with the loss of a major logistics contract over missed deadlines that were the fault of a partner contractor, and with the loss in value of one of the real estate properties after its anchor tenant left. EBITDA—earnings before interest, taxes, depreciation and amortization, the key measure of a business’s operating efficiency—went negative for the group as a whole for two quarters in a row, something that had not happened in the holding company’s entire fifteen-year history. The board of directors seriously discussed selling one of the three business lines to keep the group afloat.

The company got through that acute period, and by every outward sign got through it well: it cut costs without panicked mass layoffs, renegotiated some of its contracts on new terms, and carried out a painful but necessary restructuring of the logistics business. Three quarters after the bottom, EBITDA was back in positive territory; two quarters after that, it exceeded its pre-crisis level. By the moment this chapter is about, exactly a year had passed since the quarter when the numbers finally stopped being alarming. And it was at this point, not at the height of the crisis, that Atlas’s board of directors ran into a problem no one had foreseen.

The problem was not the numbers—they were fine. The problem was that all the discipline, all the practices, all the habits born in the acute crisis had begun to dismantle themselves, by default, without a single decision anyone could point a finger at. The weekly risk meetings, which at the height of the crisis brought the heads of all three business lines together with the CFO, first moved to every two weeks, then to “as needed,” and the “need,” as it turned out, arose less and less often, because no one wanted to spend time discussing problems that formally did not yet exist. The reserve fund, built up during the crisis precisely so as never again to end up in a situation where the only way out is to sell a business line under pressure, had been two-thirds redistributed back into current expansion projects over ten months of calm operation—each individual decision to do so looked reasonable and justified, and no one consciously decided to “cancel the safety cushion.” The practice of a short weekly report on atypical signals—unusual delays, uncharacteristic behavior by major clients, deviations in the indicators—introduced at the height of the crisis, which, as people in the holding company later admitted, really did help them see the problem in real estate two or three weeks before it became obvious from the financial statements, had by the end of the year degenerated into a formal box-ticking exercise that line managers filled in from a template without really reading what they were writing.

Atlas’s CEO summed up what was happening at one of the board meetings with a phrase that later became almost a proverb inside the company: “We haven’t forgotten the lesson of the crisis. We’ve just stopped practicing it.” This precise observation pointed to the heart of the problem: the knowledge was still there, the memory of what they had been through had not gone anywhere, and everyone remembered those two quarters in the red perfectly well. But the practice—daily, reproducible, built into the calendar and into job descriptions—had fallen apart, because it rested not on procedure but on the residual emotional tension of the acute period. And tension, as is its nature, subsided over time.

It is important to stress here what makes Atlas’s story telling specifically for the topic of business resilience, not just for the topic of discipline: the international standard ISO 22316, which sets out the principles of organizational resilience—an organization’s ability to absorb shocks and adapt in a changing environment—states directly that resilience cannot be a property of a single department or a single procedure; it must be a capability that runs through strategy, culture and risk management at the same time and is governed at the level of top management. The update of this standard being prepared for 2026 goes even further and explicitly assigns responsibility for the resilience of the organization as a whole to the board of directors, not just to executive management. This integrated, board-level view is exactly what Atlas lacked: the individual practices born in the crisis existed in isolation, each held up by its own champion, and none of them was established as a duty of the board of directors specifically, rather than the personal initiative of a particular CFO or head of the risk function.

A further warning sign was what happened to two new initiatives launched after the company had come out of the crisis. The industrial business started a pilot project to enter an adjacent market segment—a reasonable, well-calculated bet on diversification. The logistics business began implementing a new cargo tracking system. Both initiatives followed the standard, pre-crisis project management model: a single plan, a single budget, quarterly reporting on progress against the plan. When, at a regular board meeting, one of the independent directors who had joined the board after the crisis asked point-blank, “And what will we do if in six months it turns out that one of these two projects was a mistake?”—a pause hung in the room. There was no answer to the question, because no one had asked it in advance. A company that almost two years earlier had survived three blows at once and had learned to cope with it had returned to a model in which every bet was assumed to pay off.

Diagnosis through the cycle: what exactly got stuck

To understand what was happening to Atlas a year after hitting bottom, it helps to recall how the cycle works at the level of the organization as a whole, not just at the level of the individual. An organization, like a person, goes through the Search phase—assessing what is happening and what it threatens—through the Decision phase, through the phase of productive action and through the Reward phase, when the result is recognized and it is possible to move on to the next, slightly more difficult task. The psychologist Richard Lazarus, one of the founders of modern emotion theory, showed that what a person feels is determined not by the event itself but by how they appraise it—whether it is significant, whether it is controllable, whether there are resources to cope. The same is true for a group: how an organization responds to a risk signal depends not on the objective seriousness of the signal but on how the procedure for assessing it is designed. In the acute crisis, Atlas built such a procedure from scratch, under the pressure of necessity. A year later it turned out that the procedure had not outlived its creator—the crisis itself.

This is the place to apply an idea that the risk researcher Nassim Taleb examined in detail in his work on antifragility: an antifragile system is not one that withstands a blow and returns to its previous state, but one that grows stronger precisely thanks to repeated, completed cycles of stress and subsequent recovery. Ordinary resilience is rubber, which stretches and returns to its original shape. Antifragility is a muscle, which after a controlled load becomes stronger than it was. The key word here is “repeated”: a one-off load, even a very heavy one, trains the system only once. For a system to grow stronger in a lasting way, load and recovery must recur regularly, as a built-in practice rather than a one-time emergency. This is exactly what happened to Atlas a year later: the company had once gone through the load and benefited from it—but had not built in a mechanism for regularly repeating that load in calm, non-crisis times. And without repetition, a muscle built up once weakens over time in exactly the same way a physical muscle weakens without regular training.

The diagnosis by Atlas’s working group—made up of the CFO, the head of risk management and two independent members of the board of directors—identified three specific gaps in the cycle, each of which explained part of the overall picture of how the crisis discipline was being dismantled.

The first gap was the lack of an institutional owner of the Search phase. At the height of the crisis, the role of “the one who looks for weak signals” was played by all the managers at once, because fear made this function a priority on its own: people who objectively have reason to fear for the company’s fate notice deviations more sharply and report them more willingly than people who believe everything is under control. The neuroscientist Joseph LeDoux described the amygdala as the brain structure that detects threats quickly, before conscious reflection, and its activity rises with perceived threat. When the crisis passed and the subjective level of risk for most employees dropped back to normal, the very same mechanism that during the crisis had made people scrutinize deviations stopped working with its former force—simply because it no longer perceived the situation as threatening. This is not a malfunction or negligence on the part of particular people but a natural physiological reaction: a system designed to run only on the fuel of anxiety inevitably stalls when the anxiety goes away. This means the search for weak risk signals cannot be left resting on general background fear—it has to be assigned to a specific role and a specific procedure that do not depend on whether people feel anxious right now or not.

The second gap was the lack of a mechanism that would turn a detected signal into a decision without having to convene the crisis command center all over again. In the acute crisis, Atlas had built a short chain, literally twenty-four hours long: signal—assessment—decision—action. A year later this chain had not formally disappeared: the board of directors still existed, the finance committee still met. But the speed of getting from signal to decision had itself stretched from a day to several weeks, because the usual pre-crisis approval procedures had returned, designed for calm times rather than for a timely response to risk. The psychologist Daniel Kahneman, who spent many years studying how people and organizations make decisions under uncertainty, showed that a significant share of errors in judgment is explained not by the bias of particular people but by “noise”—the unpredictable variability of decisions that the decision-making procedure itself generates if it is not built rigorously enough. The crisis version of Atlas was set up so that noise in decisions was minimal: a short chain, clear authority, a minimum of intermediate approvals. The peacetime version brought back the old level of procedural noise—and with it, the old slowness.

The third gap, the subtlest and the most destructive in its long-term consequences, was the disappearance of the Reward phase for the practice of vigilance itself. While the company was in the acute crisis, maintaining risk discipline needed no separate recognition—the company’s survival was the reward, visible every day. When the crisis passed, this natural reward disappeared, and no one created an artificial reward, built into management practice, for continuing the same discipline. A manager who kept meticulously filling in the weekly report on atypical signals in peacetime received neither recognition nor time set aside specifically for this work—but did lose time they could have spent meeting the usual, measurable, rewarded KPIs (key performance indicators), the metrics on which managers are paid bonuses and evaluated at the end of the year. The psychologist Charles Carver and his co-author Michael Scheier, who studied the mechanism of behavioral self-regulation, showed that behavior rests on a feedback loop, and without a signal of progress toward the goal, effort weakens and over time dies away, even if a person rationally understands its benefits—emotional reinforcement is stronger than rational understanding in determining what a person will do regularly and what they will drop at the first opportunity. The report on atypical signals did not die out at Atlas because someone decided to cancel it—it died out because no one reinforced its continuation once the natural crisis reward had disappeared.

Putting the three gaps together, Atlas’s working group formulated a diagnosis that the board of directors later adopted as the starting point for rebuilding the holding company’s entire risk management: the company had not lost the knowledge gained in the crisis—it had lost the institution that was supposed to hold on to that knowledge in peacetime. The survival practices rested on the personal memory and residual anxiety of particular people, not on a corporate governance structure that would survive a change of mood, a change in the leadership team and the ordinary human tendency to relax once the immediate danger has passed.

Here it is worth drawing a distinction that the working group considered central to the entire subsequent plan of action: resilience and antifragility are not the same thing, and mistaking one for the other partly explains what happened to Atlas over the past year. Resilience is the ability to withstand a blow and return to the previous state; this is exactly what the holding company successfully demonstrated during the crisis itself. Antifragility is something more: the ability to become stronger precisely because of the stress endured, not merely to survive in spite of it. A year earlier, Atlas had proved its resilience. But a year later it turned out that resilience does not turn into antifragility automatically, simply from the fact of having passed a test—that takes a separate, deliberate step: turning a one-off successful experience of going through the cycle into a constantly reproduced practice. Without this step, the company remains where it was before the crisis—just with a fresher and more painful memory, which will inevitably fade.

The solution, step by step

Atlas’s board of directors made a decision of principle that set the logic of the entire subsequent restructuring: the cycle of risk management and recovery was to be moved from the mode of a one-off emergency measure into the mode of a permanent, mandatory corporate governance practice—with assigned roles, a calendar and accountability at the level of the board of directors, not at the level of the personal initiative of individual managers.

Step one. Establishing a standing resilience committee of the board of directors with a direct right to demand information from all three of the holding company’s business lines. Until then, risk issues had been considered within the finance committee as one of many agenda items, always after more urgent, routine topics. The new committee received a narrow but clear mandate: not to manage the business lines’ operations, but to systematically track risk signals across the whole group, assess the state of the reserves and once a quarter report directly to the board of directors, bypassing the filter of operational management. A telling precedent for such a step in global practice is the reorganization of safety oversight at Boeing after the crashes of its 737 MAX aircraft: back then it was a standing safety committee reporting directly to the board of directors that became the structural way to entrench what had previously rested on a one-off reaction to tragedy. Atlas applied the same logic not to flight safety but to the resilience of the business as a whole: to turn the response to a crisis from a one-off decision into a permanent management structure.

Step two. Giving the reserve fund the status of an inviolable requirement rather than whatever free cash is left over. The board of directors set a rule: the liquidity reserve created to cover a shock in any of the three business lines cannot be directed to current expansion projects by any individual decision below the level of the board of directors, and any exception requires a separate vote recorded in the minutes, with an explanation of why the temporary deviation is justified and when the reserve will be restored. This eliminated precisely the mechanism through which Atlas’s reserve had melted away by two-thirds over ten calm months: previously, each individual decision to use it looked locally reasonable, because no one saw the cumulative effect of the safety cushion gradually eroding. Separately, the company introduced capital allocation along the lines of the barbell strategy that Nassim Taleb described in his work on antifragility: the bulk of the holding company’s resources stays in a reliable, conservatively managed core business, while a separate, small share, delineated in advance, is deliberately set aside for risky, high-potential initiatives such as the industrial business’s pilot entry into a new segment. The point of this split is that it rules out the intermediate, most vulnerable state—when a company risks its whole core for the sake of a single bet without realizing it.

Step three. Introducing a mandatory quarterly “weak signals review,” written into the board of directors’ calendar, regardless of whether there is any visible cause for concern at the moment. Previously, discussion of atypical deviations came up on an ad hoc basis, when something had already gone wrong. Now, four times a year, regardless of the group’s current financial health, the board of directors spends a fixed amount of time reviewing the resilience committee’s report: which atypical signals were noticed over the quarter across all three business lines, which of them turned out to be false alarms and which deserve further monitoring. The key difference from pre-crisis practice is that it is mandatory regardless of the emotional backdrop: the decision to discuss weak signals no longer depends on whether anyone in the company feels anxious right now; it is fixed by procedure as firmly as the approval of the annual financial statements.

Step four. Splitting the management of new initiatives into two different modes—“bet” and “operation”—with different sets of checkpoints. After that episode of the board’s silence in response to a direct question about a fallback plan for the new projects, the holding company introduced a rule: at launch, any new initiative with a budget above a certain threshold is explicitly classified either as an “operation”—an expansion of an already proven, predictable model, subject to the usual quarterly reporting cycle—or as a “bet”—a move into uncertainty, for which predefined checkpoints are mandatory; if they are not reached, the initiative is automatically stopped or reconsidered rather than continuing by inertia simply because the money has already been spent. Both pilot projects discussed at that memorable board meeting were retroactively reclassified as “bets,” and for both, checkpoints and a pre-agreed wind-down scenario in case of failure were retroactively defined—something that had not existed at the moment of the independent director’s awkward question.

Step five. Introducing recognition and material reinforcement for maintaining risk discipline—not just for operating results. The board of directors built a separate, substantial component into the annual performance evaluation of the heads of the business lines: not only meeting operating and financial targets but also the quality, completeness and timeliness of their work with risk signals in their area of responsibility. This directly addressed the very gap in the Reward phase that had killed the weekly report on atypical signals: now, meticulously filling in this report no longer competed for a manager’s time with the indicators they are paid bonuses for—it had itself become one of those indicators.

Step six. Institutionalizing regular drills—controlled simulation of failure in peacetime, without a real crisis. Inspired by the documented practice at Netflix, where chaos engineering—the deliberate, regular injection of controlled failures into live systems—became a permanent, built-in process rather than a one-off experiment, Atlas introduced its own version at the level of the business, not just technology: twice a year, the resilience committee, together with the management of the business lines, holds a closed-door drill simulating one of the plausible blows—the loss of a major client, the collapse of a key contract, a sharp drop in demand in one of the business lines—and assesses how quickly and with what resources the team can actually respond, with no consequences for the real business. These regular drills became Atlas’s practical way of artificially reproducing the very load whose regular repetition Nassim Taleb wrote about as a necessity: training the muscle of antifragility without waiting for a real blow.

The board of directors codified all six steps not in the minutes of a single meeting but in amendments to the committees’ charters and to the annual planning system—that is, it moved them out of the category of management decisions by the current leadership team and into the category of corporate governance structure that will outlive particular people in particular positions. This is the precise definition of institutionalization as opposed to a one-off measure: the decision lives not in the memory of the people who made it but in documents and procedures that bind those who come after.

The result

Below are illustrative, composite figures typical of a diversified holding company of this size; they do not describe any specific existing company but show the order of magnitude achievable over a year and a half to two years of systematic work on building the resilience cycle into the ongoing practice of corporate governance.

Eighteen months after the resilience committee was launched, EBITDA for the Atlas group had grown by about twenty-two percent compared with the pre-crisis level—not as a direct payoff from the protective measures themselves (by definition they are not supposed to generate income in calm times), but as the combined result of more disciplined capital allocation and faster closing of loss-making areas within the portfolio of initiatives. The liquidity reserve fund stabilized at a level covering four to five months of the group’s operating expenses in the event of a complete halt in cash flow from any one business line—versus about a month and a half of coverage at the point when the fund had almost been dissolved into current projects.

The average time from detecting an atypical risk signal to making a decision on it fell from six to eight weeks, typical of the year after the crisis, to five to seven days—comparable to the speed of response the company had at the height of the acute crisis itself, but now achieved not at the cost of constant nervous strain but through a built-in procedure. Of the two pilot initiatives reclassified as “bets,” one—the industrial business’s entry into an adjacent segment—was wound down in an orderly way at the second checkpoint, with minimal losses, instead of continuing by inertia for another year or year and a half, as would have happened under the old project management model; the second, the logistics cargo tracking system, passed all its checkpoints successfully and was moved to the status of a regular operation.

The regular drills—eight held over two years, two a year for the group as a whole plus separate ones for each business line—identified and eliminated three real vulnerabilities in the supply chain and contractual obligations before they had a chance to turn into a real crisis: in particular, one of the drills exposed the logistics business’s excessive dependence on a single transport provider on a key route—a vulnerability that by its nature was an almost mirror-image repeat of what had once triggered the holding company’s original crisis, only this time it was closed by diversifying suppliers over three months, as planned, rather than in an emergency over three weeks under the pressure of missed deadlines.

One more outcome deserves a note of its own, one the board of directors had not originally built into the program’s goals: the independent directors who had joined the board after the crisis noted that the quality of strategic discussions at meetings had noticeably improved—discussion of new initiatives began to include an explicit conversation about the failure scenario and the way out of it, something that had been practically absent from the company’s pre-crisis culture. The institutionalized habit of asking “what if this doesn’t work?” spread from the narrow topic of risk management to the holding company’s general decision-making culture.

An important caveat, typical of stories like this: institutionalizing the resilience cycle does not mean that Atlas is insured against the next serious blow—reserves, signal discipline and regular drills reduce the likelihood of repeating precisely the combination of mistakes that led to the last bottom and shorten the response time to a new combination, but they do not eliminate the very possibility of a crisis. The point of institutionalization is not that there will be no more crises. The point is that the company does not start learning to go through the cycle anew each time, as if the previous crisis had never happened, but enters the next blow already trained, with working structures, not just with the memory that things were once hard.

Telling, too, is how the very agenda of Atlas’s board meetings changed—not only their content but also their emotional tone. The CFO who had been at the origin of the restructuring later described the difference this way: before, the conversation about risks arose suddenly, in the middle of discussing completely different issues, and always carried a tinge of anxiety, because it came up precisely when something had already gone wrong. Now the conversation about risks is a known, expected agenda item, prepared for as calmly as the discussion of the quarterly financial statements. It is this change of tone—from unexpected and alarming to planned and businesslike—that is perhaps the most accurate outward sign that the cycle has truly stopped being a one-off measure and has become part of ongoing management: routine here is not a sign of indifference but a sign that the practice is embedded so deeply that it no longer needs a surge of emotion to set it in motion.