Sunk Cost Fallacy at Work: How to Quit the Right Things

The sunk cost fallacy at work is the reason you are still funding a project everyone privately knows has failed. It is the pull to keep going because of what you have already spent — the months, the budget, the reputation — rather than because of what the next month will actually return. The fix is not more discipline. It is a deliberate stop decision: a short, repeatable ritual that asks what you would choose if today were day one and the previous spend did not exist.

I have spent most of my working life around teams that are excellent at starting and finishing, and almost untrained at stopping. We build systems for throughput — limits, batches, blocks, sprints — and none of them ask whether the thing moving through the system still deserves to exist. Throughput on the wrong project is just waste at a higher speed.

This article is about workplace decision-making, not health advice. Where it touches stress and burnout, it describes population-level research findings, not predictions about you, and it is no substitute for a conversation with a qualified professional.

Key Takeaways

• The sunk cost fallacy at work is continuing to invest because of past spend rather than expected future return — the money, hours, and credibility already gone are, by definition, unrecoverable.
• Research distinguishes two different decisions that popular advice constantly blurs: utilization decisions (do I use the thing I already paid for?) and progress decisions (do I keep funding it?). Only the second one compounds.
• Escalation of commitment is one of the ten behavioural biases Bent Flyvbjerg identifies as most consequential in project management, and surveys put roughly 30 to 40 percent of software projects somewhere on the escalation spectrum.
• The most reliable countermeasure is a kill criterion written down before the work starts, when nobody is defending anything yet.
• Stopping early is a sustainability lever, not just a productivity one: demolition accounts for more than 90 percent of the 600 million tons of construction and demolition debris the EPA counted in a single year.
• Persistence is sometimes correct. The test is whether the reason you are continuing points forward or backward.

What the Sunk Cost Fallacy at Work Actually Is

A sunk cost is any resource you have already spent and cannot get back. The fallacy is letting that unrecoverable spend influence a decision that should be made purely on what happens next. Economists have a blunt version of the rule: when deciding whether to invest further in a venture, disregard what you have already invested.

In project management the same behaviour has a more formal name. Bent Flyvbjerg, who has spent decades studying why megaprojects overrun, defines escalation of commitment as “the tendency to justify increased investment in a decision, based on the cumulative prior investment, despite new evidence suggesting the decision may be wrong.” He ranks it among the ten behavioural biases that matter most in project work, alongside the planning fallacy, anchoring, and strategic misrepresentation — drawn from a field where behavioural scientists have now catalogued more than 200 distinct biases.

The proverbs got there first. Throwing good money after bad. In for a penny, in for a pound. What the research adds is that the bias is normally invisible from the inside. In its original formulation, escalation of commitment is unreflected and non-deliberate — people do not know they are doing it, which is precisely why willpower is the wrong tool.

The Experiment That Named It

The classic demonstration is almost comically small-scale. Hal Arkes and Catherine Blumer, in their 1985 study in Organizational Behavior and Human Decision Processes, sold season tickets to the Ohio University theatre. Some buyers paid the full fifteen dollars; others were randomly handed a two-dollar or seven-dollar discount at the counter. Same seats, same plays, randomly assigned price.

The full-price group attended more performances. Not better performances — simply more of them. The seats were identical and nothing about the value of the plays had changed. Only the size of the hole in their wallet differed, and that hole was doing the deciding.

Arkes and Blumer’s interpretation is the part that transfers to work: the psychological engine is the desire not to appear wasteful. Which is a genuinely strange trap, because the thing you do to avoid looking wasteful — sitting through the play, shipping the doomed feature — is the actual waste.

The Distinction Almost Everyone Misses: Using Versus Continuing

Here is the part I wish someone had told me a decade earlier, and which nearly every article on this topic skips.

A meta-analytic review published in Business Research pooled 98 effect sizes on the sunk-cost effect and made a point of separating two decision types that had been quietly lumped together for years. Utilization decisions ask whether to consume something you already bought — the gym membership, the conference ticket, the software licence. Progress decisions ask whether to commit more resource to something already underway.

The review found clear evidence that the sunk-cost effect is real across both, but that its size and the things that moderate it depend on which decision you are making. In particular, the authors found support for the idea that the effect is attenuated by time in utilization decisions. The unused ticket stops nagging. The pull fades.

Progress decisions do not work like that, and the asymmetry is the whole problem. A utilization mistake costs you once and then decays. A progress mistake renews itself every single week, because each week’s fresh spend becomes next week’s sunk cost and next week’s justification. The fallacy is self-feeding in exactly the situation where it is most expensive.

The same review also reported that older adults were less likely to fall prey to the effect than younger ones, which I read less as a comment on age than as a comment on how many times you need to watch a doomed project cross the finish line before the lesson lands.

How Common Is the Sunk Cost Fallacy at Work?

Common enough that it shows up in the base rates rather than the anecdotes. Flyvbjerg’s overview assembles outcome data from 2,062 projects — the most comprehensive set of base rates in project management scholarship — and identifies base-rate neglect as a primary reason projects underperform. Teams do not compare themselves to what happened to everyone else who tried this. They compare themselves to the version of the plan they are already committed to.

In software specifically, the numbers are more direct. Mark Keil, Joan Mann and Arun Rai, in their empirical analysis in MIS Quarterly, found that somewhere between 30 and 40 percent of software projects exhibit some degree of escalation — continuing after negative feedback has already arrived. Software is particularly prone to it because progress is intangible and genuinely hard to measure, so the negative feedback is always arguable.

Flyvbjerg’s catalogue of case studies makes the same point at civic scale: Expo 86, the Shoreham nuclear power plant, Denver International Airport. Each one, in his phrase, its own version of driving into the blizzard.

That blizzard image comes from a thought experiment worth carrying around. Two friends have tickets to a game a long drive away. A snowstorm arrives. The higher the price they paid for the tickets, the more likely they are to drive into the storm — investing more time, more money, and more actual physical risk, on the strength of money that is already gone.

Why Your Brain Fights the Stop Decision

Three forces show up every time I have watched a team refuse to kill something, and none of them are stupidity.

The first is self-justification. Stopping is a public admission that the earlier decision was wrong, and in most organisations the person who has to make the stop decision is the same person who made the start decision. You are asking someone to file the evidence against themselves.

The second is that the cost of continuing is spread out and the cost of stopping arrives all at once. Another quarter of quiet underperformance is invisible. A cancellation notice is a meeting. Humans reliably prefer diffuse pain to concentrated pain, even when the diffuse pain is larger in total.

The third is that the bias can be weaponised deliberately. Flyvbjerg quotes the film director Elia Kazan describing his tactic openly: get actors under contract, build sets, expose negative, and “so get the studio in deep.” Once enough money has been spent, halting becomes an irretrievable loss not only of money but of face. Somebody in your organisation has read that playbook, whether or not they have read that book.

Underneath all three sits ordinary cognitive load. Stopping something is a high-stakes decision, and high-stakes decisions get deferred to a mythical later when you will somehow have more capacity — which is its own well-documented failure mode I have written about in decision fatigue at work.

How I Run a Stop Decision

I do this quarterly, on a calendar block, for everything I own that has been running more than six weeks. It takes about forty minutes. The point of scheduling it is that a stop decision should never be an emotional event triggered by a bad week — it should be a boring recurring appointment where the answer is usually “keep going.”

Ask the zero-base question

One sentence, asked out loud: if this landed on my desk today, with zero hours spent and zero money committed, would I start it?

If the answer is an immediate yes, you are done; the past spend was irrelevant either way. If the answer is no, you have just learned that the only argument for continuing is an argument about the past. That does not automatically mean stop — but it does mean every reason you are about to give needs to survive being restated in future tense.

Write the kill criterion before you start

This is the single highest-leverage habit on this list, and it costs about ten minutes at kickoff. Before any work begins, while nobody has anything to defend, write down the specific observable condition under which this project ends. Not “if it’s not working.” Something a stranger could check: if we do not have ten paying users by the end of March, we stop and write up what we learned.

You are pre-committing while you are still neutral. When March arrives you are no longer arguing about whether to quit — a decision that feels like a referendum on your judgment — you are just reading a number you wrote yourself. It converts a progress decision back into something closer to a utilization decision, which is the one research suggests we handle better.

Separate the decision from the person

Escalation is, at root, identity defence. So the practical move is to make stopping a low-status-cost act. In my own notes and in every review I run, the framing is that the project failed a test the project agreed to in advance — not that a person was wrong. Teams that publicly reward a clean early stop get far more of them, and the ones that quietly punish it get expensive zombies instead.

Harvest the salvage before you close the file

Stopping badly wastes the one thing the project actually produced: information. Before you shut anything down, spend thirty minutes writing what you now know that you did not know at the start, what assumption broke, and which components are reusable. That write-up is often worth more than the deliverable would have been, and it is the difference between a stop and a loss.

The residue of an abandoned project — the open loops, the half-finished threads, the mental tabs you never closed — will keep drawing on your attention if you never formally close it out, which is the mechanism behind the Zeigarnik effect at work. A stop decision that is never written down is not really a stop.

The Waste You Only See After You Stop

This is where the sunk cost fallacy at work stops being a productivity topic and becomes an environmental one, and it is the reason the subject belongs on this site rather than in a business-school reader.

A project pushed past its expiry date does not just burn hours. It converts those hours into material. Servers get provisioned. Prototypes get tooled. Fit-outs get built. Flights get taken to review work that should never have reached review. Every week of escalation has a physical invoice that nobody puts on the project’s balance sheet.

The scale of that invoice is easier to see in construction than in knowledge work, because construction leaves rubble. The EPA estimated 600 million tons of construction and demolition debris generated in the United States in 2018 — more than twice the country’s municipal solid waste for the same year — of which just under 144 million tons went to landfill. Demolition, not construction, accounts for more than 90 percent of that stream. Things get built, and then they get taken down.

Knowledge work generates the same pattern without the rubble to shame us into noticing. The abandoned platform still ran on real electricity. The archived data still sits on real disks in a real building with real cooling. I make this argument at greater length in digital decluttering for focus, but the short version is that the greenest week of a doomed project is the week you cancel it.

There is a neat symmetry here with the case for work in progress limits. Limits stop you starting too much. A stop decision stops you carrying too much. You need both, because a system that only constrains intake will eventually fill up with things that were once a good idea.

What to Say When You Do Not Have the Authority to Stop

Most people reading this cannot cancel anything. That is the honest situation for the majority of employees, and advice that ignores it is just theatre. What you can do is make the stop case well enough that the person who can cancel it has something to act on.

Three things make that case land. Lead with the forward number, never the backward one — what the next quarter costs and returns, with the spent budget explicitly set aside as irrelevant. Offer a smaller alternative rather than a void, because “stop” reads as defeat and “stop this, start that” reads as reallocation. And name the opportunity cost in concrete terms: the specific other work these two people would be doing instead.

What does not work is being right in private. Quietly doing the minimum on a project you believe is dead is its own trap — it looks like compliance, costs you the same hours, and produces exactly the performed effort I described in productivity theater at work. If you have concluded the thing should stop, the useful version of that conclusion is written down and handed to someone.

There is a wellbeing dimension worth naming plainly. Working month after month on something you believe is pointless, while defending it in public, is a recognisable form of sustained workplace strain. The World Health Organization describes burn-out as an occupational phenomenon — not a medical condition — resulting from chronic workplace stress that has not been successfully managed, with three characteristic dimensions: energy depletion or exhaustion, increased mental distance or cynicism about the job, and reduced professional efficacy. That middle dimension is the one escalation feeds most directly. If that pattern sounds familiar and it is not shifting, the useful next step is a person, not a process, and I have written a fuller starting point in signs of burnout and how to recover.

When Persistence Is Actually the Right Call

I want to be careful here, because “kill your darlings” advice taken literally produces its own pathology: teams that abandon everything at the first hard month and never learn anything that takes longer than a quarter to learn.

Persistence is correct when the reason for it points forward. Benefits that genuinely arrive late — infrastructure, trust, brand, skill — look identical to escalation from the outside for a long time. Learning curves are real; the fourth attempt is often cheap precisely because the first three were expensive. And reliability has compounding value: an organisation known for finishing gets offered better work, and that reputation is a future asset, not a past cost.

It is also worth separating a stop decision from a bad estimate. Most projects that feel like failures at month three are not failing — they are simply arriving where the base rates always said they would, and I have written about why that gap is structural in time estimation at work. Running late is not evidence of the sunk cost fallacy. Running late and being unable to state what would make you stop is.

The test I use is a single question about tense. Ask why you are continuing, and listen to whether the answer is in the past or the future. “We have already put eight months into this” is the fallacy. “The next three months are the cheapest they will ever be, and here is why” is a reason. Same project, same eight months, completely different decision.

Summary

The sunk cost fallacy at work is the pull to keep investing because of what you have already spent, and it is expensive mostly in progress decisions — the ones that renew themselves weekly — rather than in the one-off utilization decisions that popular advice tends to use as examples. Roughly a third of software projects show some escalation, and escalation sits on Flyvbjerg’s list of the ten behavioural biases that do the most damage to project work.

The practical defences are unglamorous. Schedule stop decisions as boring recurring appointments rather than crisis responses. Ask the zero-base question. Write the kill criterion at kickoff, while you are still neutral. Make stopping a low-status-cost act so people will do it out loud. Harvest the salvage before you close the file.

And notice what stopping actually buys. Not only the hours, but the servers, the materials, the travel, and the quiet strain of defending work you no longer believe in. The greenest and often the most productive week of a failing project is the week somebody finally ends it.

Frequently Asked Questions

What is the sunk cost fallacy at work in simple terms?

It is continuing to spend time, money, or effort on something because of what you have already spent, rather than because of what you expect to get back from here on. The already-spent resources cannot be recovered whatever you decide, so they are not information about the future. A clean decision compares only the cost of the next step against the value of the next step.

How do I tell the difference between the sunk cost fallacy and healthy persistence?

Listen to the tense of your own reasoning. If the justification is about accumulated investment — how long it has taken, how much has been spent, how bad it would look to quit — that is the fallacy. If the justification describes specific future returns and why the remaining cost is worth them, that is persistence. The same project can pass this test one quarter and fail it the next.

What is a kill criterion and how do I write one?

A kill criterion is an observable condition, agreed at the start of a project, that ends the project when it occurs. Good ones are specific enough for someone outside the team to check without interpretation: a date, a number, a threshold. Write it before any work begins, because that is the only moment when nobody involved has anything to defend, and revisit it only with the same seriousness you would apply to any other commitment.

Why is stopping a project so much harder than starting one?

Because the costs are shaped differently. Continuing spreads its cost thinly across future weeks where nobody has to look at it, while stopping concentrates the cost into one uncomfortable conversation. Stopping also usually requires the person who made the original decision to publicly revise it, which research on escalation of commitment links to self-justification. None of that is irrationality in the insulting sense; it is a predictable response to how the incentives are arranged.

Does the sunk cost fallacy apply to time as well as money?

Yes, and at work time is usually the larger stake. The research literature developed mostly around monetary sunk costs, and studies of sunk time have produced a more mixed picture than studies of sunk money. But the practical mechanism is the same in any workplace: hours already worked cannot be un-worked, so they belong in the write-up rather than in the decision.

Can escalation of commitment be used against me deliberately?

It can. Flyvbjerg documents the tactic explicitly, including a film director’s account of committing enough contracts and construction early on that cancellation would have meant an unrecoverable loss of both money and face. The defence is structural rather than personal: insist that approval gates re-evaluate the remaining spend on its own merits, and treat any argument that leans on how deep you already are as a signal to look harder rather than to commit faster.

Does quitting a project actually have an environmental benefit?

Ending work that will not be used avoids the material and energy it would otherwise consume — server capacity, prototypes, shipped samples, travel, and eventually disposal. The clearest illustration comes from the built environment, where EPA figures show demolition accounting for more than 90 percent of construction and demolition debris generation. Knowledge work produces a less visible version of the same thing, which is precisely why it goes uncounted.

How often should I review projects for escalation?

I use a quarterly block for anything running longer than six weeks, which is frequent enough to catch drift and infrequent enough that it does not turn into anxious churn. The important part is that it is scheduled in advance rather than triggered by a bad week, so that a bad week is evaluated against a standing criterion instead of a mood.