The short version
You’ve put money, time, or years into something. It isn’t working. The argument for continuing is that you’ve already put money, time, or years into it.
Those resources are gone. They’re gone if you continue and they’re gone if you stop — that’s what makes them sunk. A cost that is identical under every option cannot possibly tell you which option to choose. The only live question is what the next pound or the next year buys, and the past has no vote in it.
Where the name comes from
“Sunk cost” is straight economics — an expenditure already incurred and unrecoverable. The fallacy is the habit of counting it anyway.
Its other name is the Concorde fallacy, after the Anglo-French supersonic programme that both governments kept funding for years after the commercial case had gone. The name is worth keeping and worth handling carefully, because the history is more awkward than the label.
The 1962 agreement was a treaty, and a remarkably short one — seven articles, setting out how the work, the costs and the proceeds would be split. It said nothing whatever about either side leaving. No exit, no notice period, no penalty, no procedure. So when Harold Wilson’s incoming government tried in 1964 to renegotiate its way out, there was nothing to invoke, and the French declined to reopen it. That left Britain choosing between carrying on and breaking a treaty outright. Britain carried on.
Which makes Concorde a better story about a trap than about the fallacy it named. What mostly held the two governments in was the cost of getting out — and a cost of getting out is a forward-looking cost, the legitimate kind. Use the nickname, because it’s the one everybody knows. Just don’t use it as the proof.
What it looks like
Two years in, the platform still doesn’t do the one thing it was for.
We can’t stop now. We’ve got two years in it.
The film is dull and there are ninety minutes left.
I paid for this, I’m watching it.
The course isn’t going anywhere I want to go.
I’m three years in. I have to finish.
Why it doesn’t work
Run the arithmetic and it collapses immediately. Two options, continue or stop. The two years appear on both sides of the ledger, identically, because they’ve already happened. Cancel them out and you are left with a straight comparison of what each future costs and returns — which is the comparison you should have been making, and the one the two years were preventing.
So why is it so hard? Because stopping isn’t neutral. It’s an admission, and specifically it’s an admission about your own judgement: the earlier decision was wrong, and the spend was wasted. Continuing lets you keep the loss theoretical for a little longer, and there is a real, if temporary, comfort in that which has nothing to do with the project.
That’s also why it hardens in groups. “We’ve come too far to stop now” is rarely a forecast. It’s a way of protecting everyone who signed off on the original decision, which is usually the people in the room deciding whether to continue. The bigger the past investment, the more senior the people implicated by admitting it, and the harder the thing becomes to kill — exactly backwards from how it should work.
This is not the sunk cost fallacy
- Being nearly finished. If the remaining cost is small and the payoff still exceeds it, carry on. That’s correct marginal reasoning. The fallacy is about the past spend counting, not about ever continuing anything.
- An asset you actually still have. If three years bought you expertise, a codebase, or a licence, the three years are still sunk — they don’t come back either way. But the thing they bought is real, and it makes the next year cheaper. That lower forward cost is what belongs in the calculation. The three years still don’t.
- Commitments that genuinely bind. Contracts, promises and reputation have real future consequences. “We said we would” can be a legitimate forward-looking reason, so long as it’s the consequence doing the work and not the embarrassment.
- Refusing to churn. Abandoning things the moment they get hard has its own failure mode, and it’s at least as expensive. Persistence isn’t the fallacy; persistence justified solely by prior spend is.
The test is a clean-slate question, and it’s brutal because it works: if you arrived today, with no history here and nothing already spent, and somebody offered you this exactly as it stands — would you take it? If the answer is no, ask one follow-up before you act on it: is there anything that would cost you to stop — a penalty, a promise, an obligation that survives the decision? That’s a real forward cost and it belongs in the sum. If there isn’t, then the only thing keeping you in is a number that can’t be recovered either way.
How to call it out
- “Forget what it’s cost. What does the next six months buy?”
- “If we were starting from scratch on Monday, would we choose this?”
- “The money’s gone in both versions of this meeting. What’s different between them?”
And when you’re on the receiving end, or arguing with yourself: the honest reframe is that stopping doesn’t waste what you spent. What you spent was already spent. Continuing is how you spend it twice.