Every long project has a day when someone could have stopped it cheaply and did not. The day passes without comment. Nobody marks it on a calendar. By the time anyone looks back, the bill has grown and the reasons for continuing have changed. The trap usually looks like a habit of carrying on, not a dramatic blunder.
How the trap forms
Start with the definition. A sunk cost is money, time or effort already spent that cannot be recovered whatever you decide next. The fallacy is letting that spent amount count as a reason to spend more. The next dollar or hour should be judged on what it will produce. What was spent last year cannot change that.
Other forces feed the same behaviour, and they are worth keeping apart. Emotional investment is one. Internal politics is another. Fear of reputational damage is a third, since ending a project can look like confessing the original decision was wrong. The APM blog post on open project culture lists the sunk cost fallacy alongside politics and reputational fear as reasons underperforming work stays alive. It also notes that projects tend to launch with fanfare and end without it.
These forces overlap in practice, but they call for different fixes. A team that only needs permission to stop has a politics problem. A team that cannot stop because of what it has already spent has a sunk cost problem. Most teams have some of each.
A worked example: Meta and the metaverse
The APM post points to Meta. It reports that Reality Labs has received more than $60bn of investment since 2020 for VR and AR work, and that the division posted a $4.2bn loss in Q1 2025 alone. These figures come from that blog post. I have not checked them against Meta's own filings, so treat them as reported, not audited.
The post's argument is that Meta was slow to pivot away from the metaverse while AI drew capital and attention elsewhere. That is the sunk cost reading, and it is plausible.
There is a fair case on the other side. A long-term bet can look wasteful for years and still pay off. Early spending on an uncertain technology buys the option to be present if it matures, and that option has value even when the current numbers are ugly. Stopping also costs something. Teams scatter, learning is lost, and a later restart is more expensive than continuing would have been. The APM post does not make this case, and the outcome is not known. The useful test is not whether the losses were large. It is whether anyone could say, in advance, what evidence would have justified stopping.
Judgement, evidence and who gets heard
A second APM piece on cognitive security argues that projects often fail because of poor judgement under pressure, discounted inconvenient evidence, or deference to a confident voice. Committed teams do all three easily. Bad news is the first thing to be explained away.
The piece describes cognitive security as protecting human judgement from distortion, overconfidence and over-reliance on flawed information. It also says most risk registers cover cyber, financial and operational risks but few include risks to decision quality inside governance. One concrete detail: it reports that 70% of project professionals say their organisation uses AI, up from 36% in 2023. The post as summarised here gives no date for the newer figure, so read it as a recent snapshot. It reports too that the State of Project Management 2026 report found only half of projects are deemed successful. Neither number measures sunk cost decisions directly. They show how much judgement is being asked of people who may not be trained for it.
Setting the stop rule before you need it
The APM sources describe the problem but give no method for deciding when to stop. What follows is editorial advice, not a tested procedure from them. The common thread is to decide before commitment builds, when stopping is still cheap to discuss.
- Write kill criteria at the start. Name the evidence, the date and the threshold that would end the project, and record who owns the call.
- Run a premortem. Ask the team to imagine the project has failed a year from now and list the most likely reasons. Add those reasons to the kill criteria.
- Ask the fresh-start question. If we had no history here, would we begin this today, at this cost, with what we now know?
- Get an independent review. Someone with no stake in the earlier decisions should check progress against the criteria.
None of these removes the case for continuing. They make it explicit. A team that wants to keep going can still do so, but it has to say what has changed since the criteria were written.
Two things to carry away
First, keep the ledger honest. Money and effort already spent belong in the record, but they should not be weighed against the next decision. Only expected future costs and benefits count, and a blank column for those is a warning sign.
Second, give judgement a place on the risk register. Most registers track cyber, financial and operational hazards. Adding a line for the quality of oversight, and for who is allowed to challenge a confident voice, costs almost nothing. A hurried note in the margin is a start, but it works better as a standing item.
Sources
- Three pitfalls to creating an open project culture
- Cognitive security: The project risk no one is registering
- https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQEPA30Do5QV7z878QDHr6T2NzHiRGFj21ppqoJ1uZwQShdJLfGTo0GG67SC56iLOvKj5C3GC4HAdzRYsvIT-Reb4XS72TkBqvo_aIfdNpnwWJ80k77W-geJ30-YiZXQaMnKdy9aYapRtiSTqKmtQtVcJR9ARW6r-oe2NorPyQPCRa-jna9XO8IYuhiys7ErUBUGajhd_j3cioSGainSv5ZWodJd6kA3B9UvhoE=
- https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQEpC72Qa049L37uSvbu_BmG2L-CSb-jh8xlPaaDdCMQb32jBwIocjtkwMfI86Mglu1OcRS4bOeZjMWk8tWA2aUiHm05602rrgc3ssRiu7p129XnkxJ_HeeS3vTCXWwrCZBPMLS1CZ3JGNLkcPSw67ilIUjJDnKKESEMN4u7SWmNoEz90BKI74GUt-4vfT8dY2XbPa4mzhD9cA==
- https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGZZcYuGHvRGNxpKTwVaJpnfJWJ2lNQbFhceLkWaDy84pFDLreE7v_pkjUXdTDik7ZSAVP8m68XfTt0xJeVx8duQKs_feCMvUpbfBI0iP9pjPf6MYBN4b-YNQevEX_mD4qYYiQbaEmUeDuMSSb6KR_9riR4DhhwaqY_VSKON86gYmbTl4cM34f46D7i-7oVkRozITOmpJDfwf9UjekV7ZkCuIWLr3qdpCKlPAJQ
- The Green Book (2026) - GOV.UK
- https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQHbC4Fj8R5-chKMRjpZvPClHVIl_q-ZK-yrjPzLO4CF9Z3uuAzcCVV-klKtnclxhTN4OZ_J4gqxCtPDjT-17sKyMgRy1pcRZDiN1sTk1KD6HCDvb__C6atdWLgd9PkKCcEb
Marcus Adeyemi-Holt is a house writer for Deeperpectives, where he examines how ordinary routines reveal larger patterns in work, memory, and place. He draws on years of reporting on municipal records and oral histories to trace the quiet forces that shape everyday life.
