Why the Sunk Cost Fallacy Hits Harder When You Built the Thing

The standard sunk cost advice is short enough to fit on a sticky note: the money's gone either way, so decide from here, not from what you already spent. I believe that advice. I also notice it barely works on me when the sunk cost isn't money — it's something I built. There's a second, meaner version of the bias that the textbook explanation doesn't cover, because it isn't really about resources at all.

The textbook version is about resources. This one is about identity

Classic sunk cost is a resource problem: you spent $400 on a conference ticket, the conference turns out to be a waste, and you go anyway because otherwise the $400 was “for nothing.” It's irrational, but it's a clean, nameable irrationality — you're weighing a number that should be irrelevant to a decision that should only be about what happens next. Most people can be talked out of it in one sentence, because once you say it out loud, the mistake is obvious even to the person making it.

The version that shows up when you built the thing yourself doesn't respond to that sentence, because the cost you're protecting isn't the hours you spent. It's a claim about your own judgment. Killing something you made isn't just walking away from sunk effort — it's agreeing, in public or in private, that you were wrong to start it, wrong to keep going, or both. That's not a resource calculation anymore. That's a threat to how you see yourself, and threats to self-image don't yield to “the time's already spent, so ignore it.” Of course it's not just time. It's evidence, and you're the one on trial.

How to tell which version you're actually in

The two feel different once you know to check. Money-based sunk cost feels like reluctance — a mild, talk-yourself-out-of-it pull toward not wasting what's spent. Identity-based sunk cost feels like defensiveness. You don't feel reluctant to cut the thing; you feel the need to argue for it, sometimes to people who never asked. A few tells I've learned to notice in myself:

• You get more articulate defending it as the evidence against it piles up, instead of less — you're building a case, not weighing a decision.
• You compare the option of stopping to “giving up” rather than to “reallocating,” even though those are the same action described two different ways.
• You feel a flash of something closer to shame than regret when someone suggests cutting it, which is a strange reaction to have about a spreadsheet or a feature or a plan.
• You keep it alive at a reduced, unofficial version — the zombie state — because turning it off formally is the part that actually feels like losing, more than the resources still going into it.

A test that actually works, because it removes you from the story

The standard advice — “ignore sunk costs, decide from today” — fails here because you can't cleanly separate yourself from the decision when the decision is also a verdict on you. So don't try to ignore the identity stake. Route around it instead. The question I actually use: if this exact thing, in its exact current state, landed on my desk tomorrow having been built by a stranger, would I choose to keep running it? Not improve it, not fix it — keep it, as is, starting from today.

That question does one specific piece of work: it strips out authorship. You're no longer judging whether you were right to start it, because you didn't start it in this version of the question — some hypothetical stranger did, and their judgment isn't on trial, only the thing itself is. What's left is close to the actual decision that matters, which was never “was I right,” it was “is this, right now, the best use of what it's costing.” Those are different questions and only one of them is useful.

The money version asks what you spent. The identity version asks who you'd have to admit you were. Only the first one is actually about the cost.

The honest caveat: the test can be too clean

This isn't a trick that always points the same direction, and it shouldn't. Sometimes the stranger-desk answer is yes — the thing is fine, your discomfort was really about something else, and the honest move is to stop second-guessing it and get back to work. The test is only doing its job if it's equally willing to tell you to keep going as to tell you to stop. If you find yourself running it repeatedly on the same decision hoping for a different answer, that's not the test failing — that's the identity stake reasserting itself, which is worth noticing on its own.

It also doesn't account for information a stranger wouldn't have. If you're the only one who knows a fix is one week away, or that the reason it's underperforming is a mistake you can now clearly name and correct, that's a real asymmetry and the test should account for it — a stranger inheriting the project wouldn't know that either, so give the version of them that knowledge before you ask the question. The point isn't to erase context. It's to erase the part of the context that's about your ego rather than the thing's prospects.

What changed once I started asking it about my own decisions

I run this more than I'd like to admit, on things much smaller than a whole product decision — a template I kept using past the point it was working, a process I set up that everyone routes around now instead of following. The pattern I keep finding is that the stranger-desk answer is “no, cut it” far more often than my actual behavior would suggest, and the gap between those two numbers is basically a direct measurement of how much identity, rather than judgment, was doing the deciding.

That gap is also, I think, a decent chunk of why cutting things at a company gets slower over time rather than faster, even as the company gets better at everything else. Early on nothing has your fingerprints on it long enough to feel like a verdict on you if it goes. Later, more of the surface area is stuff you specifically built, argued for, and shipped — and every one of those is now a small identity claim sitting quietly in the product, waiting to make the next cut harder than it needs to be.

Try it on the thing you're currently defending

Pick the project, habit, or decision you've been quietly protecting from scrutiny lately — you probably already know which one, because you had a small defensive reaction just reading that sentence. Run the stranger-desk version: forget who built it, hand it a clean slate starting today, and ask honestly whether you'd keep it. If the answer's yes, good, that was worth confirming and you can stop relitigating it. If the answer's no, the sunk cost was never really the hours. It was the story you'd have to tell yourself about having spent them.