- Description
- The inside of a nuclear power plant cooling tower
- Source
- Jakob Madsen jmadsen · original page
- Licence
- CC0 Public domain
Escalation of commitment
When a project is going badly, the temptation is to bet more rather than acknowledge that the initial choice needs to be revisited.
In brief
- What: after bad news, commitment increases (budget, staffing, timeline, support) instead of being reduced or redirected.
- Why: feeling responsible for the initial choice pushes people to justify it; giving up means publicly acknowledging a mistake.
- Detect: every review asks for “a little more” resources or time, and the person who launched the project is the one deciding to continue it.
- Avoid: stop criteria written in advance, staged funding, an external decision-maker, the question “what if we started from scratch?”
Definition#
Fictitious illustrationEscalation of commitment (escalation of commitment) is the tendency to continue, and even intensify, a course of action that is delivering poor results, rather than correcting or abandoning it. Psychologist Barry Staw described it in 1976 in a paper whose title quotes a Pete Seeger song: “Knee-deep in the big muddy.”
It is close to Sunk costs, but not identical. A sunk cost refers to the weight of past spending in a decision. Escalation describes a dynamic: decision after decision, commitment grows, because each piece of bad news is treated as a reason to “hold on” or “go all in” rather than stop.
It is often the same people who launched the project and decide what happens next. Staw showed that this personal responsibility is exactly what amplifies the effect.
The mechanism#
The literature (Staw and Ross, 1987; Brockner, 1992) distinguishes several forces acting together:
- Self-justification: the person who chose a direction tries to show their choice was right. Continuing gives them a chance to prove it.
- Hope of turning it around: people assume poor results are temporary and that extra effort will get everything back on track (see Unrealistic optimism).
- Filtered information: people notice what supports continuation and downplay the rest (see Confirmation bias).
- Social and institutional pressure: a group that approved the project does not want to contradict itself; exit costs in reputation or relationships are added to the spending (see Groupthink).
- Real exit costs: contract penalties, layoffs, image with the client or the public. These costs are sometimes legitimate, but they can overshadow comparison with the cost of continuing.
Staw asks business students to play the finance director of a fictional company and allocate research funds to one of two divisions. In a second step, they are told that the chosen division performed poorly, and they are asked to allocate new funds. Those who had made the first choice themselves allocated, on average, more to the struggling division than those shown the same history as a decision made by someone else. Responsibility for the initial choice increases commitment afterward.
Where it appears#
In project management#
- Project behind schedule: at every steering committee, extra staff, another month, or additional budget is added, without reassessing expected value. The initial scope, meanwhile, is never questioned.
- Underperforming vendor: they get another chance, then another, because replacing them would require acknowledging the initial choice was poor.
- Overlong redesign or migration: “we’re 80% done,” without anyone able to say what remains; people are added to the team, then added again.
- Sponsor tied to the project: they defended the case before leadership and can no longer backtrack without contradicting themselves; bad news is reframed before being escalated (see The illusion of control).
- Technical debt: workarounds are added around a poorly designed component rather than replacing it, because replacing it would mean admitting its design was flawed.
In management#
- Struggling employee: support keeps increasing (training, accommodations, role changes) beyond what their progress justifies, out of attachment to one’s own hiring decision.
- Team or internal initiative: a program launched by an executive is renewed every year with more resources, while results stagnate.
- Failed reorganization: rather than reversing course, a second one is added “to fix the effects of the first.”
- Manager’s career decision: the person who promoted someone keeps defending that promotion long after the first signs of mismatch.
Famous cases#
The Shoreham nuclear power plant 1966-1994#
Bias at play: escalation of commitment, self-justification.
What happened. In 1965-1966, the Long Island Lighting Company (LILCO, New York State) announced a nuclear power plant in Shoreham at an estimated cost of about 75 million dollars (65 to 75 million depending on sources). Construction began in 1972; planned output was increased, safety requirements evolved, productivity was poor, and the bill rose year after year. The plant was completed in 1984, but New York State and local governments opposed its operation, especially because an evacuation plan could not be accepted. In 1985, LILCO obtained authorization for low-power testing only. In 1989, the company agreed never to operate it; the plant was sold in 1992 to the Long Island public power authority for one dollar, then dismantled. It never produced commercial electricity.
Jerry Ross and Barry Staw (1993) studied this file as a case of escalation: they show how management, shareholders, and regulators reinforced their commitment at every stage, largely because of decision-makers' past responsibility, political pressure, and the search for legitimacy.
What this illustrates
The cost was multiplied by more than sixty, without any of the successive decisions to continue ever being reset based on the plant's expected value. An important caveat: part of the increase came from inflation, increased capacity, and regulatory changes unrelated to bias, and the plant was also a major political and legal issue. The case mainly shows how the same decision-makers repeated the same commitment for more than twenty years, and why planned exit points are necessary.
Vancouver's World Exposition (Expo 86) 1986#
Bias at play: escalation of commitment, institutional and political forces.
Jerry Ross and Barry Staw studied Expo 86 in Vancouver (Administrative Science Quarterly, 1986): British Columbia remained committed to organizing the exposition even as the projected deficit grew rapidly. The authors see it as a “prototype” of escalation and suggest it starts from project-related and psychological forces, then over time becomes a more structural phenomenon—institutional and political in nature: funding continues less out of conviction than because the project, now public and committed, can no longer be stopped without visible cost.
What this illustrates
An event or project with a fixed date and image stakes can worsen the effect: the deadline makes pulling back more visible and more costly. Many “prestige” IT projects (launch expected by leadership, announcement to the client) fall into this category.
Why software projects spiral out of control 1995-2000#
Bias at play: escalation of commitment in IT projects.
Mark Keil and coauthors studied information systems projects that continue despite accumulating negative signals (Keil, 1995; Keil, Mann and Rai, 2000). They found escalation is linked in particular to the personal responsibility of the project manager or sponsor, the symbolic importance of the project, exit costs perceived as high, and the absence of stop criteria. Keil and Robey's study (1999), for its part, examines how troubled projects were de-escalated: the common factor is often the arrival of a new person or an outside perspective, which reframes the problem and proposes stopping or redefining.
What this illustrates
What gets you out of escalation is rarely one more number: it's a change in decision-maker or framing (independent review, new sponsor, formal stop point). Planning for this outside perspective at launch costs less than improvising it in the middle of a crisis.
Elf's “sniffer planes” (France) 1976-1979 ; revealed in 1983#
Bias at play: escalation of commitment (the follow-up: funding continues), after wishful thinking (the beginning: wanting to believe).
What happened. Two inventors claimed they could detect oil deposits from a plane thousands of meters underground; the state-owned oil group Elf funded them for three years for hundreds of millions of francs. The process was a hoax. After an initial contract for 400 million francs, others followed (including 500 million in June 1978), even though doubts had been raised: the amounts already committed seemed to force continuation. Contracts were not stopped until summer 1979, and the case remained secret for four more years.
The full account, with sources and limitations, is in the Wishful thinking page.
What this illustrates
Set stop criteria in advance and validation milestones assigned to people outside the project.
Bygmalion: the 2012 campaign continued despite warnings (France) 2012-2025#
Bias at play: escalation of commitment (what follows: continuing despite warnings), among other causes.
What happened. During the 2012 presidential campaign, organizing numerous rallies caused spending to skyrocket. On March 7, 2012, the campaign accountant, Pierre Godet, sent the campaign director a note estimating the projected budget at €23,150,971, i.e., €642,000 above the legal cap, while 5 out of 44 rallies had taken place; he recommended limiting spending and forbidding any additional commitments. A second warning, dated April 26, 2012, was sent to the candidate. The campaign continued. To conceal the overspending, the Bygmalion agency billed the UMP, according to court rulings, around 58 fake invoices totaling about €18 million. The Paris criminal court convicted Nicolas Sarkozy on September 30, 2021 for illegal campaign financing; the Paris court of appeal (February 14, 2024) sentenced him to one year in prison, including six months to be served, a ruling made final by the Court of Cassation on November 26, 2025. He was not prosecuted for organizing the fake invoice system, for which other parties were convicted.
Where the bias appears. The courts established that the candidate, an experienced one, had been warned in writing about the risk of overspending and that the campaign continued at the same pace, without spending controls on his part. This is the classic form of escalation of commitment: the further the campaign goes, the harder it seems to cancel already announced rallies, and each warning is treated as an obstacle to work around rather than a reason to reconsider. But the ruling does not characterize the candidate's psychology: it finds a deliberate failure to control, that is, a choice, and the fake invoice system shows that the overspending was organized and concealed by others. So at best, the bias explains only continued spending, not its concealment. Level of evidence: accepted, not proven. The warnings and continuation are established; reading them as escalation of commitment is an interpretation.
What this illustrates
When a quantified warning announces that a legal limit will be exceeded, the response must be a written decision (stop, reduce, or accept), made by someone who did not commit the spending, before the next expense.
Detection signals#
- “One last push is all it takes.”: the same phrase at every steering committee.
- Requested resources increase while observable results have not changed.
- The decision-maker is the project's promoter; no outsider has had to validate continuation.
- Indicators are changed or redefined when they become bad (new progress calculation method, new scope “to better reflect reality”).
- People talk about reputation (“it would be a disastrous signal”) more than remaining value.
- Bad news reaches higher levels softened, and doubters stay silent.
Avoidance reflexes#
- Write stop criteria before starting. Cost threshold, date, value milestone, usage indicator: “If by this date we haven't reached this value, we stop or redirect.” They act as a contract with yourself.
- Fund in stages. Each stage is a real decision point, with the possibility to stop without having to justify “losing” the total budget.
- Have someone unconnected to the choice decide. An independent committee, a new sponsor, a peer from another project: an outside perspective is the best documented remedy.
- Separate the past from the future. For each decision to continue: “If we were starting from scratch today, would we launch this project at this cost?” (see Sunk costs).
- Ask two people for the same file. One presentation “for” and one presentation “against” continuation, with the same data.
- Make exit honorable. Talk about redirection, value what you keep and what you've learned; people don't admit an error when they risk paying the personal price.
- Run a pre-mortem before each new stage. Imagine the project failed in one year: what are the causes?
With AI: detect, fix, mitigate#
AI has no reputation to protect: it can play the role of an outside decision-maker, redo an analysis without the project's history, and review successive reports to spot escalation. It still depends on the file it is given, however, and it tends to follow the framing of the question: “how can this project succeed?” calls for an answer that helps it continue, not stop.
Detect#
- Compare successive committee reports: increasing resource requests, unchanged indicators, repeated wording (“one last push”).
- Identify definition changes in an indicator (progress, scope, end date).
- Highlight arguments focused on reputation or consistency rather than remaining value.
Fix#
- Produce a “from zero” analysis: remaining cost, expected value, three options (continue, reduce, stop).
- Write the two versions of the case file: the most convincing one to continue and the most convincing one to stop.
- Prepare the communication for a strategic shift (what we keep, what we learn).
Mitigate#
- Have stop criteria drafted at launch and reviewed at each committee meeting.
- Assign a devil's advocate role before any request for additional resources.
- Generate a premortem at each funding stage.
Prompts to integrate into your practices#
1. Outside view on a resource request. Use before a committee that must decide on extra funding.
Here is the history of a project: objective, initial budget, budget spent, planned and achieved milestones, latest resource requests, and expected value: [data].
Play the role of a decision-maker who did not launch this project. Ignore what has already been spent.
1. Estimate remaining cost and duration (with a range) and the most fragile assumptions.
2. Compare, based only on future costs and benefits, three options: continue with extra funding, reduce scope, stop.
3. Point out signs in the history that suggest escalation (repeated requests, changed indicators).
4. End with a recommendation in two sentences, specifying what information is missing for you to decide.2. Write stop criteria. At the launch of a project or a funding stage.
I am launching the following project: [objective, scope, budget, deadline, expected value].
Propose five to seven measurable stop or reorientation criteria, and for each: the threshold, the verification date, the person responsible for checking it, and the resulting decision.
Distinguish what justifies stopping from what justifies only reducing scope. Do not assume any value I did not provide: indicate what you need to ask me.How to integrate them#
- At launch: prompt 2 produces a draft of stop criteria, which the team discusses and signs.
- Before each continuation committee: the project manager runs prompt 1 on their own case file, notes what is concerning, then presents it.
- Afterward: the criteria are reviewed, and any change to a criterion must be justified in writing.
Precautions#
- A flattering file produces a flattering analysis: also attach unfavorable figures and opposing views.
- Remaining cost estimates are orders of magnitude: validate them with the people doing the work.
- AI may cite nonexistent studies or cases: verify every reference.
- Do not send budgets, contracts, or personal information to an unauthorized tool.
Limits#
- Escalation is not always irrational: continuing may be the best option if remaining cost is lower than remaining value, or if stopping has its own costs (penalties, customer loss, site closure).
- Conflict of interest can mix with bias: a decision-maker may have a real interest in not stopping. The diagnosis is then more nuanced than “they are wrong.”
- The meta-analysis by Sleesman and colleagues (2012), which combines 166 independent samples, confirms that the effect exists, but shows that it depends heavily on context (decision-maker role, reputational stakes, available information, presence or absence of stop criteria). The orders of magnitude from one experiment do not transfer directly to a real project.
- Famous cases (Shoreham, Concorde, Expo 86) combine political, technical, and regulatory reasons; they illustrate the dynamic without proving a single cause.
Sources and verification#
- Staw (1976), “Knee-deep in the big muddy: A study of escalating commitment to a chosen course of action,” Organizational Behavior and Human Performance — the foundational experiment.
- Staw & Ross (1987), “Knowing when to pull the plug,” Harvard Business Review (March 1987)
- Brockner (1992), “The escalation of commitment to a failing course of action: Toward theoretical progress,” Academy of Management Review
- Sleesman, Conlon, McNamara & Miles (2012), “Cleaning up the big muddy: A meta-analytic review of the determinants of escalation of commitment,” Academy of Management Journal
- Ross & Staw (1993), “Organizational escalation and exit: Lessons from the Shoreham nuclear power plant,” Academy of Management Journal, 36(4), 701-732
- Ross & Staw (1986), “Expo 86: An escalation prototype,” Administrative Science Quarterly, 31(2), 274-297
- Shoreham Nuclear Power Plant (Wikipedia, in English) — dates (construction from 1972, tests in 1985, 1989 agreement, sale for one dollar in 1992); cost figures vary by source, from more than 5 to around 6 billion dollars.
- Keil (1995), “Pulling the plug: Software project management and the problem of project escalation,” MIS Quarterly
- Keil, Mann & Rai (2000), “Why software projects escalate: An empirical analysis and test of four theoretical models,” MIS Quarterly
- Keil & Robey (1999), “Turning around troubled software projects: An exploratory study of the deescalation of commitment to failing courses of action,” Journal of Management Information Systems, 15(4), 63-87
- Le Club des juristes (November 26, 2025), “Bygmalion case: the Court of Cassation rejects Nicolas Sarkozy's appeal” — establishes the final nature of the conviction (one year, including six months in prison, appellate ruling of February 14, 2024).
- Wikipedia, “Bygmalion case” — timeline, alerts from March 7 and April 26, 2012, fake invoices and successive decisions; secondary source, to be cross-checked against rulings.
- La DH / AFP (September 30, 2021), “Bygmalion case: sentenced to one year in prison” — reports the first-instance judgment: spending of around €42.8M for a €22.5M cap, failure to exercise oversight, no prosecution for organizing fake invoices.
See also#
Related biases#
Variants or cousins: not to be confused.
Counterweights and related reflexes#
What neutralises it.
- Les critères d'arrêt fixés à l'avance (concept or method, no page)
- The pre-mortem
- Le décideur extérieur au projet (concept or method, no page)
- The devil's advocate
