Escalation of Commitment and the Sunk Cost Trap

The Most Expensive Decision You Never Made Was the One to Continue

A health system is 18 months and $4.2 million into a $3 million EHR customization project. The original timeline was 12 months. The project is now projected to cost $6.5 million and take another 14 months. An off-the-shelf alternative would cost $2.8 million to implement from scratch, with a 10-month timeline. The rational decision is obvious: abandon the custom build and switch. The actual decision, in the overwhelming majority of cases, is to continue — because stopping means admitting that $4.2 million was wasted.

This is escalation of commitment, and it is one of the most expensive decision failures in healthcare transformation and grant-funded programs. It is not a character flaw. It is a predictable consequence of how human cognition processes loss, identity, and uncertainty — and healthcare creates nearly perfect conditions for it to flourish.

Sunk Costs: The Definition That Everyone Knows and Nobody Follows

A sunk cost is a cost that has already been incurred and cannot be recovered regardless of future action. In normative economic theory, sunk costs are irrelevant to forward-looking decisions. The only inputs that should matter are the expected future costs and benefits of each available option. Whether you have spent $0 or $4.2 million on the current path should not change the analysis of what to do next, because that money is gone either way.

This is among the most well-established findings in behavioral economics, and among the most consistently violated. Arkes and Blumer (1985) demonstrated the sunk cost effect in controlled experiments: subjects who had paid more for a theater subscription attended more performances — even bad ones — than subjects who had paid less, because the prior expenditure created psychological pressure to “get their money’s worth.” The cost was sunk. Attending a bad play does not recover it. But the feeling that not attending wastes the investment is powerful enough to override rational analysis.

The mechanism is not stupidity. It is loss aversion. Kahneman and Tversky’s prospect theory (1979) established that losses are felt approximately twice as intensely as equivalent gains. Abandoning a $4.2 million project registers psychologically as a $4.2 million loss — even though the money is already spent and continuing the project does not recover it. The decision-maker is not weighing future costs and benefits. They are weighing the pain of crystallizing a loss against the hope of eventual vindication.

This distinction matters operationally because it identifies the intervention point. The problem is not that decision-makers lack information about sunk costs. Most healthcare leaders can define the sunk cost fallacy. The problem is that knowing the definition does not neutralize the psychological mechanism. Loss aversion operates below conscious reasoning, and it is amplified by every contextual factor that healthcare transformation provides in abundance.

Escalation of Commitment: When Sunk Costs Become a Spiral

Barry Staw’s foundational research (1976) identified a pattern more dangerous than a one-time sunk cost error: escalation of commitment, the tendency to increase investment in a previously chosen course of action precisely because it is failing. This is not merely continuing to spend. It is spending more, doubling down, accelerating investment in direct proportion to the accumulating evidence that the investment is failing.

Staw’s original experiment assigned participants to allocate R&D funding to one of two divisions. When told their chosen division was underperforming, participants who had made the initial allocation invested significantly more in the next round than participants who had been assigned the initial allocation by someone else. The mechanism was self-justification: the decision-maker’s identity was bound to the choice, and admitting failure threatened that identity.

Three interlocking mechanisms drive escalation:

Self-justification (Staw, 1976; Brockner, 1992). The decision-maker who championed the original investment has a personal stake in its success that extends beyond the project itself. Terminating the project is not just writing off the investment — it is a public admission that their judgment was wrong. The higher the decision-maker’s visibility and the more personal their advocacy, the stronger the self-justification pressure. Brockner’s research established that escalation increases when the decision-maker is publicly identified with the project, when others are watching, and when organizational culture treats project failure as personal failure.

Cognitive dissonance (Festinger, 1957). Continuing to invest in a project you suspect is failing creates dissonance between the belief “I am a competent decision-maker” and the evidence “this project I chose is failing.” Dissonance resolution overwhelmingly favors reinterpreting the evidence rather than revising the self-assessment. The decision-maker finds reasons the project is actually on track, redefines success criteria, blames external factors, or concludes that the project just needs more time and resources. Each rationalization makes the next round of investment feel justified.

Prospect theory and the loss domain (Kahneman & Tversky, 1979). Once a project is failing, the decision-maker is operating in the loss domain of the prospect theory value function. In the loss domain, people are risk-seeking — they prefer a gamble that might recover the loss over a certain smaller loss. Terminating the project is a certain loss (the sunk cost is crystallized, the project is dead). Continuing is a gamble (it might still succeed, recovering the investment and vindicating the decision). Risk-seeking in the loss domain means the decision-maker systematically overweights the possibility of recovery and underweights the probability of further loss.

These three mechanisms compound. The decision-maker who championed the project, who is publicly associated with it, and who is already in the loss domain will almost always choose to continue. The rational calculation — expected future return on remaining investment — never gets performed because the psychological deck is stacked against it.

Why Healthcare Is Maximally Vulnerable

Healthcare transformation programs create nearly ideal conditions for escalation of commitment. This is not an abstract risk. It is a structural feature of how healthcare investment decisions are made.

Long duration. Healthcare transformation programs run 2-5 years. A Medicaid waiver implementation, a population health management buildout, or a behavioral health integration initiative cannot show results in 6 months. The extended timeline means that by the time failure signals become clear, the accumulated investment is already large enough to trigger severe sunk cost pressure.

High visibility. Transformation programs are announced publicly, featured in board presentations, included in strategic plans, and often tied to leadership performance evaluations. The CEO who announced a $5 million care coordination platform at the annual board retreat cannot quietly shut it down 18 months later without a visible admission of failure. The higher the visibility, the stronger the self-justification pressure.

Personal championship. Healthcare transformation programs almost always have an executive sponsor who personally advocated for the initiative, shepherded it through budget approval, and staked professional credibility on its success. Staw’s research specifically identifies personal identification with the initial decision as the primary amplifier of escalation.

Grant funding. Grant-funded programs add a layer of external accountability that paradoxically increases escalation. Terminating a grant-funded program means returning unspent funds, reporting failure to the funder, and jeopardizing future funding relationships. The decision-maker is not just admitting failure to themselves and their board — they are admitting it to a federal agency, a foundation, or a state health department that may fund future initiatives. The reputational cost of termination extends beyond the organization.

Ambiguous metrics. Healthcare outcomes are multidimensional, lagging, and confounded. A care coordination program showing low utilization can be reframed as “still in the adoption curve.” A population health initiative showing no cost reduction can be attributed to “external factors” or “insufficient measurement period.” The ambiguity of healthcare metrics provides unlimited material for dissonance-reducing rationalization.

Every one of these conditions — long duration, high visibility, personal championship, external funding accountability, ambiguous metrics — is independently associated with increased escalation in the research literature. Healthcare transformation programs typically exhibit all five simultaneously.

The Go/No-Go Problem

At what point should a failing program be terminated or redirected?

The rational answer is straightforward: when the expected future return on remaining investment is negative, regardless of past investment. If completing the EHR customization will cost $2.3 million more and deliver a system worth $1.8 million in operational value, the project should be terminated — even if $4.2 million has already been spent. The $4.2 million is irrelevant to the forward-looking calculation.

The psychological answer is: almost never. Because termination requires a decision-maker to perform three acts that the human cognitive system resists simultaneously: crystallize a certain loss (prospect theory), admit their judgment was wrong (self-justification), and do so publicly (social accountability).

Healthcare Example 1: The EHR Customization Spiral. A regional health system began a $3 million EHR customization project to build a care management module tailored to their population health contracts. At month 18, the project is $1.2 million over budget, 12 months behind schedule, and the vendor has disclosed that two critical integrations originally scoped as “standard” will require custom development. The projected completion cost is now $6.5 million over 32 months. Meanwhile, the EHR vendor has released a standard care management module — available at $2.8 million with 10-month implementation. The rational analysis is clear: the remaining cost to complete the custom build ($2.3 million) will produce a system inferior to the off-the-shelf module that costs $2.8 million but starts fresh. The CIO who championed the custom build, who convinced the board to approve the initial investment, and who has reported “progress” at six consecutive board meetings, recommends “staying the course with a revised timeline.” The board approves, because they too have endorsed the project publicly and termination feels like admitting collective failure.

Healthcare Example 2: The Grant-Funded Care Coordination Program. A federally funded care coordination program was designed to reduce emergency department utilization among high-risk Medicaid beneficiaries. The grant covers 24 months at $1.8 million. After 18 months and $1.35 million in expenditure, enrolled patient utilization has not changed meaningfully. The program manager’s analysis reveals that the enrollment model assumed patients would self-refer based on outreach, but the target population does not engage with mailed materials or phone-based outreach. Rather than reconsidering the fundamental model — perhaps the intervention should be embedded in primary care visits rather than operating as a standalone program — the program manager doubles down on outreach: more mailers, more phone calls, a community health worker to do door-to-door visits. The investment in outreach intensity is a classic escalation pattern. The first 18 months of investment in the standalone model would appear wasted if the model were redesigned, so the model is preserved and the inputs are increased. The remaining 6 months and $450,000 are spent amplifying a delivery mechanism that the data have already shown does not work.

De-Escalation Strategies

Escalation is not prevented by awareness. Decision-makers who know about the sunk cost fallacy still commit it, because the mechanisms operate below the level of conscious reasoning. De-escalation requires structural interventions that change the decision environment.

Pre-commitment to kill criteria. Before the project starts, define the specific, measurable conditions under which the project will be terminated or restructured. “If the project exceeds 130% of budget at any milestone, the project is paused for independent review.” “If enrolled utilization has not decreased by 10% at month 12, the delivery model is redesigned.” Kill criteria work because they separate the termination decision from the continuation decision. The decision to terminate was made at a time when no sunk costs existed and no self-justification pressure was active. Executing the kill criteria at month 12 is implementing a prior decision, not making a new one.

Rotating decision-makers. The person who decides whether to continue a project should not be the person who decided to start it. Staw’s research directly supports this: participants who did not make the initial allocation decision showed no escalation. Assign continuation reviews to a different executive, a different committee, or an external party. The new decision-maker has no identity investment in the prior choice and can evaluate the forward-looking case on its merits.

External review. Bring in an evaluator who has no relationship with the project, the decision-maker, or the funding source. External reviewers are immune to the self-justification and social accountability pressures that drive internal escalation. Grant programs that require independent mid-point evaluations are structurally de-escalating, whether or not they recognize it.

Separating the investment decision from the continuation decision. Frame every continuation review as if it were a new investment decision. “If we were starting from scratch today, with no prior investment, would we choose this approach?” If the answer is no, the only reason to continue is sunk cost.

The Pre-Mortem: Prevention Before Escalation Takes Hold

Gary Klein’s pre-mortem technique is the most effective preventive tool against escalation because it operates before the project starts — before sunk costs accumulate, before identity attaches, before the loss domain activates.

The method is simple. Before launch, the project team imagines it is 18 months in the future and the project has failed. Each team member independently writes down the most likely reasons for the failure. The team then aggregates and discusses the results.

The pre-mortem works through two mechanisms. First, it surfaces risks that optimism bias and groupthink suppress during normal project planning. When the instruction is “this project will succeed — what could go wrong?”, team members self-censor to avoid appearing negative. When the instruction is “this project has failed — what caused it?”, the psychological permission to name risks is dramatically higher. Second, the pre-mortem creates a documented record of anticipated failure modes that can be converted directly into kill criteria and monitoring triggers. If the team identifies “vendor cannot deliver the integration on time” as a pre-mortem failure cause, that becomes a milestone trigger: “if the vendor misses the integration milestone by more than 30 days, initiate independent technical review.”

In healthcare transformation, the pre-mortem should be a standard gate in project initiation — conducted after the project plan is drafted but before funding is committed. The output is a set of named failure scenarios, each with a corresponding kill criterion and monitoring indicator. This converts abstract risk awareness into operational decision infrastructure that functions when escalation pressures are at their highest.

Warning Signs

• The project narrative has shifted from “on track” to “making progress despite challenges.” This linguistic shift marks the transition from execution to rationalization. When project updates start explaining why delays are understandable rather than reporting measurable progress, escalation is already underway.
• Budget increases are approved in increments rather than evaluated as cumulative totals. Each individual increase seems modest. The total — which is the number that matters — is never stated in the approval conversation. This is salami-slicing escalation: no single decision feels like a large commitment, but the cumulative effect is a project far beyond its original scope and cost.
• The original champion is still the sole voice in continuation decisions. If the person who started the project is the only one arguing for its continuation, and no independent evaluation has been conducted, self-justification is the operating mechanism, not evidence.
• “We’ve come too far to stop now” appears in any form. This is the sunk cost fallacy stated as an argument. The distance already traveled is precisely the information that should be excluded from the forward-looking decision.
• Metrics have been redefined or supplemented since launch. If the original success criteria would show failure, and new criteria have been introduced that show progress, the project has not improved — the measurement has been adjusted to reduce dissonance.

Product Owner Lens

What is the human behavior problem? Decision-makers continue investing in failing healthcare transformation programs because prior investment creates psychological pressure to persist, even when the forward-looking case is negative.

What cognitive or social mechanism explains it? Loss aversion makes termination feel like a larger loss than it rationally is. Self-justification ties the decision-maker’s identity to the project’s success. Prospect theory shifts behavior toward risk-seeking in the loss domain, favoring the gamble of continuation over the certainty of termination. Cognitive dissonance drives reinterpretation of failure evidence.

What design lever improves it? Pre-commitment to kill criteria before investment begins. Separation of the initial investment decision from the continuation decision. Rotation of decision-makers for continuation reviews. External evaluation at defined milestones. Pre-mortem analysis at project initiation to surface risks and define triggers.

What should software surface? (1) Cumulative investment tracking against original budget, displayed as a single number — not incremental approvals — at every review point. (2) Kill criteria status: green/yellow/red indicators for each pre-committed threshold, updated automatically from project data. (3) Milestone-based program health dashboards that compare actual performance against original success criteria — not revised criteria. (4) Decision audit trails that record who approved each continuation and what evidence was presented, making the escalation pattern visible retrospectively.

What metric reveals degradation earliest? Budget variance rate: the speed at which actual spending is diverging from planned spending, measured at each milestone. A project that is 10% over at month 6 and 25% over at month 12 has a worsening variance rate, which is a stronger escalation signal than the absolute overage at any single point. Secondary indicator: the number of scope changes or success criteria modifications since project launch, which tracks rationalization activity.

Integration Hooks

OR Module 6 (Scenario and Stress Testing). The pre-mortem and kill criteria described on this page are operational implementations of scenario analysis. OR Module 6 provides the formal framework for constructing named downside scenarios and measuring plan resilience. Human Factors Module 4 provides the psychological rationale: without pre-commitment to decision criteria before the project starts, the cognitive mechanisms of escalation will prevent rational evaluation once the project is underway. The connection is bidirectional — scenario analysis identifies what could go wrong, and kill criteria convert those scenarios into enforceable decision triggers. A stress-tested project plan that does not include pre-committed kill criteria has done only half the work: it knows what could fail but has not solved the harder problem of ensuring that decision-makers will act on that knowledge when escalation pressures are active.

Public Finance Module 4 (Milestone and Program Execution). Milestone-based disbursement — releasing grant funds only upon demonstrated achievement of defined deliverables — is a structural de-escalation mechanism. It works because it externalizes the continuation decision. When the funder controls the next tranche of funding and releases it only upon milestone achievement, the decision to continue is no longer solely in the hands of the project champion. The funder, who has no personal identity investment in the project’s approach, evaluates forward-looking evidence at each gate. This is Staw’s rotating decision-maker principle implemented through funding structure rather than organizational process. Grant programs that front-load disbursement sacrifice this structural protection and leave continuation decisions entirely within the escalation-prone internal decision environment.