Class 3

Thinking in the Burning Building

How experienced leaders actually make decisions under extreme time pressure — and why classical rationality gets it wrong

The Myth of Rational Choice Under Fire

For most of the twentieth century, the dominant model of good decision-making was the rational choice model: identify the problem, generate a comprehensive set of options, evaluate each option against defined criteria, select the optimal one. This model, rooted in economic theory and refined in operations research, works beautifully in boardrooms, planning committees, and academic examinations. It is how we teach MBAs to think. It is how we structure strategic plans. And it is almost entirely useless when the building is on fire.

Herbert Simon saw the problem decades before anyone studied fire commanders. In Administrative Behavior, Simon (1947) introduced the concept of bounded rationality — the recognition that human decision-makers cannot evaluate all alternatives because they lack the time, the information, and the cognitive capacity. Instead, they satisfice: they set an aspiration level for what constitutes an acceptable outcome and choose the first option that meets it. Good enough, fast enough. This was heresy in an era that worshipped optimization, but Simon won a Nobel Prize for it, because he was describing how humans actually behave rather than how economists wished they would.

Yet even Simon's model implies a degree of deliberation — setting criteria, scanning options, evaluating whether each meets the threshold. What happens when you have neither the time to scan nor the cognitive bandwidth to evaluate? What happens when the decision must be made in seconds, the information is ambiguous or contradictory, and the consequences of getting it wrong are measured in human lives?

That question drove a young cognitive psychologist named Gary Klein into fire stations, military command posts, and neonatal intensive care units in the early 1980s. What he found there overturned nearly everything the decision sciences thought they knew.

---

Recognition-Primed Decision Making: How Experts Actually Think

Klein's original study was elegantly simple in design and revolutionary in findings. He and his colleagues interviewed 26 experienced fireground commanders — professionals with an average of 23 years of service — about 156 critical decisions made during actual fires (Klein et al., 2010). The research question was straightforward: when you face a life-or-death decision under extreme time pressure, how do you choose what to do?

Classical decision theory predicted that commanders would generate several courses of action and compare them. They did not. In 80 to 90 percent of the decisions Klein studied, commanders reported considering only a single option. They did not compare. They recognised (Klein, 1998).

The model Klein developed from these findings — the Recognition-Primed Decision (RPD) model — describes a three-stage process that experienced professionals use under pressure:

1. Situation recognition. The decision-maker rapidly categorises the current situation by matching it against patterns stored from prior experience. A fireground commander arriving on scene does not see a building on fire — she sees this kind of fire: a ventilation-controlled basement fire in a balloon-frame structure, or a wind-driven high-rise fire with extension to the floor above. Each pattern carries with it a set of expectancies, relevant cues, plausible goals, and typical actions.
2. Mental simulation. Having recognised the pattern, the decision-maker mentally simulates a single course of action: if I do this, will it work? She runs the scenario forward in her mind, looking for problems. If the action holds up under mental simulation, she executes it. If it breaks — if she imagines a point of failure — she modifies the action or, less commonly, recognises a different pattern and simulates again.
3. Action. The decision-maker commits. The entire process, from recognition to action, can take seconds.

This is not guessing. It is not recklessness. It is what Klein (2008) calls naturalistic decision making — the study of how experienced people make decisions in real-world settings characterised by time pressure, high stakes, ambiguous information, and dynamic conditions. The NDM framework emerged precisely because laboratory studies of decision-making — which typically gave subjects unlimited time, clear options, and defined probabilities — were describing a world that crisis leaders never inhabit.

The critical insight of RPD is that expert intuition is not mystical. It is compressed experience. A fireground commander with 23 years of service has seen thousands of fires. Those experiences are stored not as explicit rules but as patterns — perceptual configurations that bundle together cues, expectations, and actions. When a new situation arrives, the commander's brain does what brains do extraordinarily well: it matches the incoming information to a stored pattern, often before the commander can articulate why. Klein calls this "seeing the invisible" — experienced professionals literally perceive features of a situation that novices cannot, because their perceptual systems have been trained by years of pattern exposure.

---

When Cognitive Load Breaks the Machine

Recognition-primed decision making is powerful, but it is not invulnerable. It requires cognitive resources — specifically, it requires that the decision-maker's working memory be available for pattern matching and mental simulation. When that working memory is overwhelmed, the entire system degrades. And it does not degrade gently.

Research on cognitive load consistently demonstrates that decision quality does not decline in a smooth, linear fashion as demands increase. Instead, performance holds relatively steady across a range of increasing load, and then drops sharply — a threshold collapse. Allen and colleagues (2014) found that when participants were placed under cognitive load by memorising an eight-digit number while making decisions, their ability to extract basic information remained intact, but their capacity to optimise choices — to think strategically — was severely suppressed. The implication is stark: under high cognitive load, you can still read the gauges, but you can no longer figure out what they mean together.

Deck and Jahedi (2015) extended these findings, showing that cognitive load increases risk aversion, reduces numerical reasoning, and makes decision-makers more susceptible to anchoring — fixating on the first piece of information they encounter, even when it is irrelevant. Under load, the analytical system that might catch errors or question initial impressions is effectively taken offline, leaving the faster, more automatic system to operate without supervision.

For crisis leaders, this threshold effect has two critical implications. First, you must manage your own cognitive load. Every additional demand — a ringing phone, a side conversation, an unresolved question held in working memory — pushes you closer to the threshold. Second, and perhaps more importantly, you must manage the cognitive load of your team. A leader who piles simultaneous demands on a subordinate during a crisis is not merely being inconsiderate; she is systematically degrading the quality of decisions across her entire operation.

The Practical Signatures of Overload

How do you recognise cognitive overload in yourself or others? The research identifies several warning signs that experienced leaders learn to watch for:

• Tunnel vision: Fixation on a single information source or a single hypothesis, with failure to check alternatives or scan the broader environment.
• Shed tasks without noticing: Quietly dropping responsibilities or monitoring duties that feel less urgent — not by deliberate triage, but by unconscious neglect.
• Regression to familiar actions: Executing well-practised routines even when the situation has changed and those routines are no longer appropriate.
• Communication breakdown: Increasingly terse or absent communication — the overloaded person literally does not have the bandwidth to talk and think simultaneously.
• Emotional flattening or irritability: Either a noticeable loss of affect (robotic responses) or a sharp increase in frustration, both of which signal that executive cognitive resources are depleted.

---

Three Decisions That Illuminate the Model

Sullenberger on the Hudson: Recognition Under the Clock

Return to the cockpit of Flight 1549. The National Transportation Safety Board's investigation (2010) revealed something remarkable about Sullenberger's decision-making. In post-accident simulations, pilots who attempted an immediate return to LaGuardia — without any delay for assessment or decision-making — were able to land safely. But when the simulations added a realistic 35-second delay for the pilots to assess the situation and decide on a course of action, every single attempt to return to the airport failed. The plane crashed into buildings.

Sullenberger's decision to ditch in the Hudson was, in Klein's framework, a textbook recognition-primed decision. He did not compare LaGuardia versus Teterboro versus the Hudson using weighted criteria. He recognised the situation — dual engine failure at low altitude over an urban area — and mentally simulated the return to LaGuardia. In his mind, the simulation broke. He could see they would not make it. The Hudson appeared not as an optimised choice but as the only course of action that survived mental simulation. Critically, he also made several expert decisions that overrode standard procedure: he activated the auxiliary power unit immediately (ahead of checklist sequence) and selected flaps 2 instead of the standard flaps 3, a choice that reduced drag and extended glide distance. These were not analytical calculations. They were the products of decades of accumulated flight experience expressed as pattern recognition.

Wag Dodge at Mann Gulch: Creative Insight at the Edge of Death

In Chapter 1, we examined the Mann Gulch disaster as a case of organisational collapse. Now we return to it through a different lens: the decision-making of foreman Wag Dodge in the final minutes before the fire overtook his crew.

As Maclean (1992) reconstructs it, Dodge realised that the fire had jumped the gulch and was racing uphill toward his crew at a speed that made escape to the ridge impossible. He had perhaps ninety seconds. What Dodge did next was, in Klein's analysis, one of the purest examples of creative decision-making under lethal time pressure ever documented. He stopped running, bent down, and lit a match. He set fire to the grass in front of him, creating a small burned-out area. Then he lay down in the ashes of his own fire and let the main fire burn over him.

No one had ever done this before. There was no training for it, no protocol, no prior pattern to recognise. Dodge invented the escape fire in the moment. Klein has analysed this decision as a case where RPD's normal pattern-matching mechanism was unavailable — the situation was genuinely novel — and the decision-maker's deep understanding of the domain (how fire behaves, what it needs, what it leaves behind) enabled a creative leap. Dodge's crew, lacking this depth of understanding, saw only a madman lighting a fire in the path of a fire. They ran past him. Thirteen of them died.

Fukushima Daiichi: When Analytical Thinking Paralysed Action

The Fukushima Daiichi nuclear disaster of March 2011 offers a contrasting lesson. When the earthquake and tsunami struck, operators at the plant faced an unprecedented situation: the complete loss of electrical power, instrumentation, and cooling systems across multiple reactor units simultaneously. Emergency management plans had never contemplated this scenario (National Research Council Committee, 2014).

In the critical early hours, decision-makers at Fukushima attempted to use analytical processes — verifying plant status, consulting procedures, seeking authorisation through the chain of command — under conditions that demanded faster, more adaptive action. With no reliable instrument readings, operators could not confirm reactor status, which meant they could not match the situation to their trained procedures, which meant they could not act. The procedural framework that was designed to ensure careful decision-making became a paralysis mechanism. Hours were lost in attempts to gather information that simply was not available, while conditions inside the reactors deteriorated beyond recovery.

The lesson is not that analytical thinking is bad. It is that analytical thinking deployed in conditions that require recognition-based or creative action is a form of mode mismatch — and mode mismatch kills.

---

Choosing the Right Decision Mode

The central practical question of this chapter is not "which decision mode is best?" It is "which decision mode is appropriate right now?" The answer depends on a dynamic assessment of four factors:

1. Time available. When time is measured in seconds, recognition-primed decisions are not merely preferred — they are the only option. As available time increases to minutes and hours, analytical approaches become feasible and sometimes necessary.
2. Novelty of the situation. RPD works when the situation matches a known pattern. When the crisis is genuinely unprecedented — a novel virus, a never-before-seen system failure, a conflict with no historical parallel — pattern matching may lock the decision-maker onto the wrong template. These situations demand creative thinking, which requires a deliberate pause, even when every instinct screams to act.
3. Expertise of the decision-maker. RPD requires a large library of stored patterns, built through years of relevant experience. A novice cannot recognise patterns that have never been encoded. Novices in crisis situations are generally better served by structured analytical tools — checklists, protocols, decision trees — precisely because they lack the experiential base for reliable intuition.
4. Stakes and reversibility. When a wrong decision can be corrected, fast recognition-based action with subsequent adjustment is efficient. When a wrong decision is catastrophic and irreversible — launching a military strike, administering a lethal drug dose — the higher cost of error may justify slower analytical verification, even under time pressure.

The most dangerous moments in crisis leadership occur at the boundaries between these modes — when a situation looks familiar but is actually novel (leading to confident pattern matching onto the wrong template), or when novelty is present but time pressure triggers an automatic reversion to known patterns. Expert leaders develop what we might call mode awareness: the metacognitive ability to monitor not just the crisis but their own cognitive process, asking "Am I in the right mode for this situation?"

---

Satisficing Under Pressure: Good Enough, Fast Enough

Simon's concept of satisficing gains new urgency in crisis conditions. In a crisis, the search for the optimal decision is not merely impractical — it is actively dangerous, because the time consumed searching for "best" allows the situation to deteriorate past the point where any good option remains. The satisficing leader sets a clear threshold: this action will be adequate to prevent the worst outcome and preserve options for future adjustment. Then she acts.

This requires a psychological discipline that many high-achievers find deeply uncomfortable. Leaders who have been rewarded throughout their careers for finding the best answer, the elegant solution, the thoroughly analysed recommendation, must learn to override that training in crisis. The perfect is genuinely the enemy of the good when the building is burning. A good-enough decision made now is almost always superior to an optimal decision made too late.

But satisficing has a cost. It requires cognitive resources to set appropriate aspiration levels, to monitor whether the "good enough" decision is actually performing adequately, and to adjust when it is not. In prolonged crises — those lasting days, weeks, or months — the cognitive resources that make satisficing possible are themselves eroded by fatigue, stress, and the cumulative weight of decision after decision. When those resources are finally exhausted, decision-makers do not satisfice; they default. They fall back on the most automatic, most habitual response available, regardless of whether it fits the situation. This dynamic — the erosion of satisficing capacity over time — is a critical vulnerability in extended crises, one we will examine in detail in Chapter 6.

Practical Strategies for Protecting Decision Quality

Understanding the cognitive science of crisis decision-making is valuable only if it translates into action. Drawing from the research reviewed in this chapter, several strategies emerge for leaders who must maintain decision quality under conditions designed to destroy it:

• Reduce extraneous cognitive load ruthlessly. Every unnecessary piece of information, every avoidable interruption, every unresolved ambiguity consumes working memory. Crisis leaders must aggressively strip away non-essential demands — on themselves and their teams — to preserve capacity for the decisions that matter.
• Build pattern libraries before the crisis. RPD only works if the patterns are already stored. Realistic scenario training, after-action reviews, case study analysis, and exposure to diverse crisis types all expand the library of recognisable situations. The time to build intuition is before you need it.
• Designate a mode monitor. Assign someone on the team the explicit role of monitoring the decision process itself: Are we in the right mode? Are we analytically deliberating when we need to act? Are we pattern-matching when we should be questioning our assumptions? This metacognitive function is one of the first casualties of cognitive overload in the primary decision-maker.
• Create structured pauses. Even in fast-moving crises, brief structured pauses — a ten-second "everyone stop and state what they see" — can interrupt fixation, surface missed information, and reset cognitive processes that have narrowed dangerously.
• Accept satisficing explicitly. Make the satisficing threshold visible and stated: "We are looking for an action that prevents reactor breach for the next two hours. We are not looking for a permanent fix right now." Making the aspiration level explicit prevents both under-ambition and the paralysing pursuit of perfection.