Sunk cost fallacy in action: doubling down on failing projects to justify past investment. Barry Staw's 1976 study gave MBA students the role of a financial VP allocating R&D funds between two divisions. After the first round, they received performance data showing one division had failed. The rational response is to shift resources to the performing division. The actual response: students who had made the original allocation to the failing division invested significantly more in it the second time. They escalated. Students who inherited someone else's failing allocation did not — they cut it. Same data. Same decision. The difference was personal responsibility. The person who made the original bad bet doubled down to justify it.
Vietnam. Concorde. Countless corporate projects. The pattern repeats at every scale. The Vietnam War is escalation of commitment played out across a decade and 58,000 American lives. Each troop increase was justified by the prior troop increase — withdrawal would have meant the previous soldiers died for nothing, so more soldiers were sent to make the previous sacrifice meaningful. The logic is circular and self-reinforcing: the worse the outcome, the greater the investment required to justify the prior investment.
By the mid-1970s, both the British and French governments knew the supersonic jet would never be commercially viable. The development costs had ballooned past every estimate. The economic case for cancellation was overwhelming. The political case for continuation was stronger: billions had already been spent, national prestige was at stake, and no politician wanted to be the one who admitted the programme was a mistake. They kept funding it. The plane flew for 27 years. It never turned a profit. Economists call sunk-cost-driven escalation "the Concorde fallacy." Yahoo acquired Tumblr for $1.1 billion in 2013. Within two years, the platform was declining by every metric that mattered — user engagement, advertiser interest, cultural relevance. Yahoo's response was not to write down the acquisition and redeploy the capital. It was to invest further in Tumblr's "turnaround." By 2019, Verizon (which had acquired Yahoo) sold Tumblr to Automattic for roughly $3 million. The $1.1 billion was gone. Years of management attention was gone. The opportunity cost of what Yahoo could have built with those resources was incalculable.
The mechanism combines three psychological forces that compound destructively. First, the sunk cost fallacy: past investment feels like it will be "wasted" if the project is abandoned, even though the investment is irrecoverable regardless of the decision. Second, ego protection: the decision-maker's identity is fused with the decision. Killing the project means admitting the original decision was wrong. Third, commitment and consistency: people who have publicly committed to a course of action will go to extraordinary lengths to remain consistent — even when the evidence has invalidated it.
Amazon's "two-way door" — reversible decisions don't trigger escalation. Bezos: "If you're not stubborn, you'll give up on experiments too soon. If you're not flexible, you'll pound your head against the wall." The distinction: stubbornness on the right bets, flexibility on the wrong ones. The antidote is structural, not psychological. Pre-commit exit criteria before investing more. The person who made the bet cannot objectively evaluate whether to continue it. Staw proved this in 1976. Fifty years of research has confirmed it.
The most dangerous feature of irrational escalation is that it masquerades as virtue. Persistence, determination, grit, commitment — the qualities that every leadership book celebrates are the same qualities that escalation exploits. The founder who "never gives up" is heroic when the persistence pays off and delusional when it doesn't. The difference between the two is visible only in the outcome. Without structural safeguards — exit criteria defined in advance, evaluation by someone without ego in the outcome — persistence and escalation are indistinguishable until it's too late.
Section 2
How to See It
Irrational escalation is operating whenever the primary justification for continued investment is the magnitude of prior investment — when "we've already spent so much" becomes the reason to spend more. The diagnostic: strip away the sunk costs and ask whether you would start this project today, from scratch, given everything you now know. If the answer is no, escalation is running the decision.
Startups & Product
You're seeing Irrational Escalation when a startup has spent eighteen months building a product that isn't gaining traction, and the founder's response to every board meeting is "we're so close — we just need six more months." The timeline never shrinks. The goalposts shift. The metrics that were supposed to prove product-market fit get redefined. The founder cannot process the possibility that the original thesis was wrong because their identity is fused with the product. The board sees a failing company. The founder sees an investment they refuse to write off.
Corporate Strategy
You're seeing Irrational Escalation when a company continues pouring resources into a failing acquisition rather than writing it down. HP's acquisition of Autonomy ($11.1 billion in 2011, followed by an $8.8 billion write-down in 2012) is the textbook case — but the escalation happened between acquisition and write-down, when HP kept investing in integration and market positioning for a product that forensic accounting would later reveal was built on inflated metrics. The executives who championed the deal could not evaluate its failure objectively because killing it would have been a public admission of their judgment failure.
Military & Government
You're seeing Irrational Escalation when a government continues a policy or programme primarily because reversing it would require admitting the policy was wrong. Infrastructure projects are chronic offenders: the "Big Dig" in Boston, originally budgeted at $2.8 billion, eventually cost $14.6 billion. At each cost escalation, the justification was the same — "we've already invested too much to stop." The justification is logically empty. The money already spent is gone whether the project continues or not. The only relevant question is whether the remaining cost is justified by the remaining benefit. Escalation prevents that question from being asked.
Personal Finance & Career
You're seeing Irrational Escalation when someone stays in a career they hate because of the years they've already invested. "I've spent eight years building expertise in this field — I can't walk away now." The eight years are gone. The only relevant question is whether the next year in this field produces more value than the next year in an alternative. The sunk time creates the feeling of waste. The escalation converts that feeling into a justification for continued investment in a path that the person, if starting fresh, would never choose.
Section 3
How to Use It
Irrational escalation cannot be defeated with awareness alone. The person escalating always has a reason — and the reason always feels rational in the moment. The defence is structural: build decision systems that detect escalation before it compounds and assign kill decisions to people who are not psychologically invested in the project's survival.
Decision filter
"For any project or investment that has underperformed expectations, apply the clean-slate test: if I were not already invested — no money spent, no time committed, no reputation attached — would I start this project today with what I now know? If the answer is no, the only question remaining is how to exit with minimum damage. The sunk cost is not a factor. It was never a factor. It only feels like one."
As a founder
Pre-commit to kill criteria before launching any major initiative. Write them down. Make them specific: "If we don't reach 1,000 active users by month six, we shut this feature down." "If the unit economics aren't positive by Q3, we pivot." The criteria must be defined before emotional investment accumulates — while you're still on the flat part of the commitment curve, before identity fuses with the project. Jeff Bezos institutionalised this at Amazon with the concept of "two-way doors" — decisions that can be reversed. The meta-principle: design every initiative as a reversible experiment whenever possible, and define the reversal trigger in advance. The Fire Phone was a $170 million failure. Amazon killed it within a year. The speed of the kill — not the avoidance of the failure — is what separates Amazon from companies that would have spent three more years trying to make it work.
As an investor
The escalation trap in investing is called "averaging down" — buying more of a losing position to reduce the average cost basis. Sometimes this is rational: the thesis is intact, the market is mispricing, and additional investment increases expected return. Often it is escalation: the investor cannot process that the original thesis was wrong, and additional investment serves to postpone the admission of loss. The structural defence: separate the "hold/sell" decision from the "would I buy today?" decision. If you would not buy this asset today at its current price with everything you now know, you should sell it — regardless of what you paid.
As a decision-maker
Assign kill decisions to people who did not make the original commitment. Staw's research is unambiguous: the person who made the decision is the worst person to evaluate whether to continue it. Their ego, reputation, and identity are fused with the outcome. They will find reasons to continue that a disinterested observer would reject. The practical implementation: create a formal review process for underperforming initiatives where the review panel excludes the original sponsors. Give the panel authority to kill. Make the criteria explicit and pre-defined. The hardest version of this: apply it to yourself. When you are the one who made the original bet, actively seek out the person on your team most likely to tell you the truth — and give them the explicit mandate to argue for abandonment.
Common misapplication: Labelling all persistence as irrational escalation. Staw's research identifies escalation when continued investment is driven by prior investment rather than by rational assessment of future returns. But sometimes the rational assessment genuinely supports continuation — the thesis is intact, the obstacles are temporary, and additional investment has a positive expected return. The founder who keeps building through a difficult market is not necessarily escalating irrationally. The diagnostic is the justification: if the argument for continuation references the future ("the market is inflecting, our retention is improving, the unit economics are turning positive"), it may be rational persistence. If the argument references the past ("we've already invested too much to stop now"), it is escalation.
Section 4
The Mechanism
Section 5
Founders & Leaders in Action
The leaders below demonstrate the discipline of de-escalation — the willingness to kill a failing initiative despite personal investment, public commitment, and the psychological pain of admitting error.
Bezos built Amazon's culture around the principle that killing failures fast is as important as scaling successes. The Fire Phone ($170 million loss in 2014) was Bezos's personal project — he championed the hardware, overrode scepticism from lieutenants, and attached his reputation to the launch. The phone failed catastrophically: Amazon took a $170 million write-down on unsold inventory within months. Bezos's response was not to escalate — not to invest in a Fire Phone 2, not to rebrand and relaunch, not to blame the market. He killed it. Publicly. And redirected the team to Echo and Alexa — which became one of Amazon's most successful product lines. Bezos: "If you're not stubborn, you'll give up on experiments too soon. If you're not flexible, you'll pound your head against the wall." The two-way door — reversible decisions that don't trigger escalation — was the structural mechanism. Amazon celebrates failed experiments because the cost of failure is bounded by design, while the cost of escalating a failure is unbounded.
Hastings demonstrated anti-escalation under maximum public pressure with the Qwikster debacle in 2011. Netflix announced it would split into two services — streaming (Netflix) and DVD-by-mail (Qwikster) — with separate websites, separate billing, and separate queues. The backlash was immediate and severe: Netflix lost 800,000 subscribers in a single quarter, and the stock dropped 77% from its July peak. The escalation temptation was enormous: Hastings had publicly committed to the split, recorded a video explaining it, and staked his strategic vision on the separation. Reversing it meant admitting — in public, with cameras rolling — that he was wrong. Hastings reversed it within three weeks. He killed Qwikster before it launched. The speed of the reversal — the willingness to absorb the reputational hit of admitting error rather than compounding the operational damage of a bad decision — is the textbook anti-escalation move. Netflix's market cap today exceeds $250 billion. The leader who couldn't de-escalate would still be running Qwikster.
Musk illustrates the razor-thin line between rational persistence and irrational escalation — and the structural mechanisms that keep him on the right side. SpaceX's first three Falcon 1 rockets exploded. Each failure consumed tens of millions. The fourth launch was the company's last chance — a fourth failure would have meant bankruptcy. The superficial reading: Musk escalated, got lucky, and the fourth rocket worked. The structural reading: Musk had pre-committed to exactly four attempts. The budget was structured for four failures. The engineering process was designed to learn maximally from each explosion. The "escalation" was not emotional doubling-down — it was a pre-defined experimental sequence with a kill criterion. Musk's discipline is in the distinction: he persists within pre-defined bounds and kills outside them. Tesla's abandoned battery-swap programme, the cancelled Model X falcon-wing door redesign, and the deprioritised Tesla Semi all reflect a willingness to stop when the evidence warrants it — despite public commitments.
Section 6
Visual Explanation
Irrational escalation creates a divergence between cumulative investment and probability of success. The rational exit point — where additional investment produces negative expected returns — passes unnoticed because each increment feels small relative to what has already been spent, and the psychological cost of abandonment keeps rising with every dollar invested.
The diagram maps the core dynamic of irrational escalation. Two curves move in opposite directions: cumulative investment (solid red, rising) and probability of success (dashed gold, falling). At the "Initial bet" stage, both are aligned. After the first setback, the curves begin diverging. The vertical dashed line marks the rational exit point — where expected return turns negative and additional investment destroys value. In a rational system, the project would be killed here. In an escalating system, it is not — the decision-maker enters the escalation zone, continuing to invest primarily to avoid acknowledging that prior investment was wasted. The curves diverge until collapse, when the loss becomes undeniable. The gap between the rational exit point and the actual exit point (collapse) is the cost of escalation — and it is always larger than the cost of the original mistake, because the escalation adds new investment on top of the irrecoverable sunk cost. The bottom panel identifies the four psychological drivers that keep the decision-maker investing past the rational exit point: sunk costs, ego protection, consistency pressure, and loss-aversion framing that makes quitting feel worse than gambling. Each driver operates independently. When all four are active simultaneously — the typical condition in high-stakes corporate decisions — the threshold for de-escalation becomes nearly unreachable.
Section 7
Connected Models
Irrational escalation is not a standalone bias. It is the compound product of multiple cognitive distortions working in concert — sunk cost reasoning providing the raw material, ego and consistency pressure amplifying the commitment, and loss aversion framing the exit as more painful than the continuation. The counter-models provide the structural tools for detecting and interrupting the cycle.
Reinforces
Sunk [Cost](/mental-models/cost) Fallacy
The sunk cost fallacy is the foundation on which irrational escalation is built. The fallacy: past expenditures — irrecoverable by definition — are treated as relevant to future decisions. Irrational escalation is the sunk cost fallacy in motion — the ongoing process of making successive decisions contaminated by prior investment. Every new round of investment is justified by the accumulated prior rounds. The sunk cost fallacy provides the flawed logic. Escalation provides the behaviour. The person who says "we can't stop now after investing $50 million" is running the sunk cost fallacy as operating procedure. The $50 million is gone whether they invest another dollar or not.
Reinforces
Loss Aversion
Loss aversion reframes the escalation decision in terms that make continuation feel safer than exit. Kahneman and Tversky demonstrated that losses are weighted roughly 2x more heavily than equivalent gains. In the escalation context: abandoning a project crystalises a certain loss (the sunk investment is admitted to be unrecoverable). Continuing the project preserves the possibility — however small — that the investment will be recovered. The loss-averse brain prefers the uncertain gamble to the certain loss, even when the expected value of the gamble is negative. This is why escalation accelerates as the stakes increase.
Reinforces
Commitment & Consistency
Cialdini's consistency principle is the social amplifier of irrational escalation. A person who has publicly committed to a course of action — announced it to a board, told investors, described it at an all-hands — experiences intense psychological pressure to remain consistent with that commitment. Reversing course creates cognitive dissonance and social penalty: the leader who changes direction is perceived as indecisive, unreliable, or incompetent. The commitment pressure raises the threshold of negative evidence required to trigger de-escalation.
Section 8
One Key Quote
"If you're not stubborn, you'll give up on experiments too soon. If you're not flexible, you'll pound your head against the wall."
— Jeff Bezos, Amazon Leadership Principles
Bezos compressed the entire escalation trap and its antidote into two sentences. Stubbornness on the right bets — the persistence that turns a contrarian thesis into a category-defining company. Flexibility on the wrong ones — the willingness to kill the Fire Phone, admit the mistake, and redirect resources to Echo. The trap: most leaders are stubborn when they should be flexible (escalating a failing bet) and flexible when they should be stubborn (abandoning an experiment before it has time to prove or disprove itself). The distinction is not in the quality of the feeling — both feel like conviction. The distinction is in the structure: pre-committed exit criteria, two-way doors that make reversal costless, and evaluation by people who didn't make the original bet.
The two-way door is the operational translation. Reversible decisions don't trigger escalation because the psychological cost of reversal is low — you're not "admitting failure," you're "iterating." The one-way door — irreversible, consequential — demands the opposite: slow down, get it right, don't trust intuition. The leader who applies this asymmetry — fast and flexible on reversible bets, stubborn and rigorous on irreversible ones — avoids both the escalation trap and the premature-abandonment trap. Bezos didn't invent the principle. He made it structural at scale.
Section 9
Analyst's Take
Faster Than Normal — Editorial View
Irrational escalation is the bias that startup culture is structurally least equipped to detect. The entire mythology of entrepreneurship celebrates persistence in the face of doubt. "Everyone told me it wouldn't work, and I kept going" is the origin story of every successful company. The identical sentence is also the origin story of every spectacular failure. The only difference is the outcome — and the outcome is determined by factors that persistence alone does not control. Survivorship bias edits the dataset to make escalation look like virtue.
The acquisition version of escalation is the most expensive form in corporate finance. When a company acquires a target for $1 billion and the target underperforms, the acquiring CEO faces a choice: write down the acquisition (admitting the purchase price was wrong) or invest more in "integration and turnaround" (escalating). The write-down hits the P&L immediately, triggers board scrutiny, damages the CEO's track record, and feeds the business press cycle. The escalation is invisible — the additional resources flow through operating budgets, the declining metrics are explained by "integration challenges," and the reckoning is deferred. By the time the write-down eventually arrives — and it almost always arrives — the total loss includes both the original overpayment and the years of additional investment poured into the escalation.
The pattern I see most often in portfolio companies: the founding team cannot evaluate its own product because their identity is the product. When a hired executive recommends killing a feature, the data is evaluated on its merits. When a co-founder recommends killing the feature they personally designed and named, the recommendation triggers an existential crisis. The ego cost of de-escalation scales with the decision-maker's personal involvement — which is why the person most qualified to build the product is often the least qualified to decide when to stop building it. The structural solution: every major initiative needs a designated "devil's advocate" — someone whose explicit role is to argue for abandonment, and whose career incentive is not tied to the project's continuation.
The de-escalation framework that I recommend to every founder: build kill criteria the same day you build the project plan. Not as an afterthought. Not as a concession to pessimism. As the most important line item in the plan. The kill criteria should specify exact metrics, exact timeframes, and exact decision authorities. "If retention drops below 20% at month six, the VP of Product — not the founding team — decides whether to continue." Imperfect kill criteria defined in advance are vastly better than no kill criteria at all — because the alternative to pre-defined criteria is post-hoc rationalisation, and post-hoc rationalisation is where escalation lives.
Section 10
Test Yourself
Irrational escalation hides inside reasonable-sounding justifications. The decision-maker always has a story about why continuation is rational — the market is about to turn, the product needs one more iteration, the team just needs more time. The scenarios below test whether you can distinguish genuine rational persistence from escalation driven by sunk costs, ego, and commitment pressure.
Is this irrational escalation?
Scenario 1
A SaaS startup has spent $4 million and 18 months building a product. Monthly active users have been flat at 200 for six months. The founder tells the board: 'We've spent $4 million getting to this point. Our engineers know the codebase inside out. We just need $2 million more and a redesigned onboarding flow to unlock growth.' The board asks what evidence supports the onboarding thesis. The founder cites three user interviews.
Scenario 2
A media company acquired a digital publishing platform for $500 million three years ago. The platform's revenue has declined 15% annually since acquisition. The CEO proposes a $100 million 'transformation programme' — new technology stack, new content strategy, new monetisation model. When a board member asks whether divestiture should be considered, the CEO responds: 'We can't sell now. The market would see it as an admission of failure, and we'd get pennies on the dollar.'
Scenario 3
A venture-backed startup has been building for two years. Six months ago, a competitor launched a similar product with better technology and raised a $50M Series B. The startup's metrics have stalled. The founder proposes pivoting to an adjacent market segment where the competitor isn't present. The pivot requires rebuilding 40% of the product and targeting a different customer profile.
Section 11
Top Resources
The irrational escalation literature connects organisational psychology, behavioural economics, decision theory, and strategic management. Start with Staw for the foundational experiments, extend through Kahneman for the prospect-theory framing that explains the risk-seeking behaviour, and apply through Bazerman for the practical organisational interventions.
The foundational experiment. Staw's MBA-student allocation study demonstrated that personal responsibility for an initial decision is the primary driver of escalation. The paper established the core finding: people who made the original commitment invested significantly more in failing projects than people who inherited the same projects. Every subsequent escalation study builds on this design.
Kahneman's prospect theory chapters explain why escalation is risk-seeking behaviour. When facing a certain loss (abandoning the project) versus an uncertain gamble (continuing with the possibility of recovery), loss-averse decision-makers prefer the gamble — even when its expected value is negative. The loss aversion and framing effects chapters provide the cognitive architecture that makes escalation feel rational from inside the decision-maker's head.
Ariely's chapter on the "cost of zero cost" and his experiments on ownership effects provide the consumer-level evidence for escalation dynamics. The endowment effect — overvaluing what you already own — is escalation's close cousin: once you own a project, a position, or a strategy, you value it more than the market does, which makes rational exit feel like irrational loss.
The most comprehensive treatment of escalation in organisational contexts. Bazerman's chapters on escalation of commitment synthesise thirty years of research into a practical framework for managers: the conditions that produce escalation, the cognitive mechanisms that sustain it, and the structural interventions that counteract it.
Klein's pre-mortem technique — first described here and later formalised in subsequent publications — is the most effective structural countermeasure to irrational escalation in the decision-science literature. By imagining the project has already failed before launch, teams create psychological permission to identify failure modes and establish exit criteria while emotional investment is still low. The pre-mortem chapter is the single most operationally useful piece of escalation-prevention advice in any book on the topic.
Irrational Escalation — cumulative investment rises while probability of success falls. The rational exit point passes silently. The decision-maker keeps investing because stopping would mean the prior investment was 'wasted.'
Tension
[Inversion](/mental-models/inversion)
Inversion — thinking backward from failure — is the conceptual antidote to escalation. Charlie Munger: "Invert, always invert." Instead of asking "how do we make this work?" ask "what would have to be true for this to fail?" The inversion flips the frame: the escalating decision-maker asks how to salvage the investment. The inverter asks what would justify abandoning it. Pre-defined exit criteria are inversion made operational — the failure conditions written down before emotional investment makes them unthinkable.
Tension
Pre-mortem
The pre-mortem is the most effective structural countermeasure against irrational escalation. Gary Klein's technique — imagining the project has already failed and working backward to identify the causes — neutralises escalation's core mechanism by making failure psychologically real before emotional investment makes it unthinkable. When a team conducts a pre-mortem at launch, the failure scenarios and exit criteria become part of the project's DNA rather than post-hoc admissions of defeat. The pre-mortem doesn't prevent escalation. It creates the decision artefacts — written exit criteria, defined failure signals — that make de-escalation mechanically easier when the evidence warrants it.
Tension
Disagree and Commit
Bezos's "disagree and commit" principle creates a structural release valve for de-escalation. When a decision is reversible (two-way door), the leader can say: "I disagree with this direction, but I commit to executing it." The commitment is to the team and the process, not to the specific bet. When the bet fails, the leader who disagreed can recommend killing it without the ego cost of admitting they were wrong — they were right to disagree. The principle doesn't eliminate escalation. It reduces the identity fusion that makes de-escalation psychologically impossible for the person who championed the original decision.
The deepest insight from Staw's research: escalation is a feature of responsibility, not of irrationality. The students who made the original allocation escalated. The students who inherited the allocation did not. Same data. Same decision. Different psychological relationship to the prior choice. This means the most powerful anti-escalation tool is rotation — regularly moving decision authority so that the person evaluating a project's future is not the same person who committed to its past.
The organisational tell that escalation is running the building: the language of obligation. Listen for "we owe it to the team," "we've come too far," "we can't let this investment go to waste." Each phrase encodes the escalation mechanism in moral language — transforming a financial decision into a duty. The team's effort is sunk. The distance travelled is sunk. The prior investment is sunk. None of these create an obligation to continue. But the moral framing makes de-escalation feel like betrayal rather than rationality. The leader who can hear this language — in their own voice, not just others' — and recognise it as escalation's vocabulary has the most important diagnostic skill in resource allocation.
