Why You Can't Let Go: The Neuroscience of Sunk Costs

The Concorde was supposed to cost seventy million pounds. By the time the British and French governments finished building it, the final cost reached roughly 1.5 to 2.1 billion pounds — more than twenty times the original budget. At multiple points during the development process, internal analyses concluded the aircraft would never be commercially viable. The numbers were clear. The market was too small, the fuel costs too high, the routes too limited. Both governments had the data. Both kept funding it.

The project became so famous for this pattern that economists named the phenomenon after it. The Concorde fallacy: continuing to invest in something because of what you've already spent, not because of what you'll get back. The sunk costs weren't just a factor in the decision. They were the decision. Each additional billion made it harder to stop, because stopping meant admitting the previous billion was wasted.

The Concorde flew for twenty-seven years. The programme never recouped its development costs.

If you're a founder, you've felt a smaller version of this. A product that isn't working but has six months of development behind it. A hire who isn't performing but cost three months to recruit. A marketing channel that hasn't converted but already consumed the quarterly budget. The rational move is obvious. The pull to keep going is stronger. And if you've ever wondered why that pull feels less like a decision and more like gravity, the answer is that it is, in a real sense, neurological. Your brain processes sunk costs on different hardware than it processes rational evaluation.

The Ski Trip You Shouldn't Take

In 1985, Hal Arkes and Catherine Blumer published a study that stripped the sunk cost fallacy down to its simplest form.

Participants imagined they had bought two ski trips. The first cost $100 and was to a resort they expected to enjoy moderately. The second cost $50 and was to a resort they expected to enjoy much more. Then they discovered the trips were on the same weekend and neither ticket was refundable.

The majority chose the $100 trip — the one they expected to enjoy less. When asked why, the answers clustered around the same logic: they'd paid more for it, and not going would feel like a bigger waste.

The participants understood they would have a worse time. They chose the worse experience anyway, because the brain doesn't process "money already spent" and "expected future enjoyment" on the same circuit. The sunk cost generates its own signal — a form of anticipated regret and loss aversion — that competes with the forward-looking evaluation and, in Arkes and Blumer's experiment, wins.

This is why even polite, well-intentioned customer feedback can't save you — the endowment effect that inflates the value of what you've built is the same circuit that makes sunk costs feel like future obligations. Neuroimaging research has confirmed that when people make decisions influenced by sunk costs, the activation patterns are distinct from the patterns associated with rational cost-benefit analysis. The brain is literally running a different computation — one that weights the past investment more heavily than the future return. Sunk cost reasoning isn't a thinking error layered on top of good decision-making. It runs on separate neural architecture.

The Invisible Drift

Sociologist Diane Vaughan coined the term "normalization of deviance" to describe what happened at NASA before the Challenger disaster. But the clearest modern example is Boeing's 737 MAX.

The MCAS system — Maneuvering Characteristics Augmentation System — was designed to correct a tendency for the 737 MAX's nose to pitch up under specific conditions. The original specification was narrow: a small correction, in rare circumstances, requiring two sensors for redundancy. What shipped was different in five ways, each one a small deviation from the original standard.

First, the scope expanded. MCAS was allowed to activate at lower speeds and in a wider range of conditions than originally planned. Each expansion was individually reviewed and approved. Second, the system was designed to rely on a single angle-of-attack sensor instead of two, eliminating the redundancy that would have caught a sensor malfunction. Third, the system was removed from pilot training manuals — Boeing argued it operated in the background and didn't require pilot awareness. Fourth, when the single sensor malfunctioned, MCAS pushed the nose down repeatedly, and the manual override required a procedure that was extremely difficult to execute under the stress and time pressure of the actual failure mode. Fifth, internal test pilots flagged concerns during development. Those concerns were documented, reviewed, and ultimately not acted upon.

No single deviation was catastrophic. Each was defensible in isolation. An engineer could explain why the scope expanded. A manager could explain why one sensor was sufficient. A training team could explain why the manual didn't mention MCAS. But the deviations accumulated, each one normalizing the next, until the distance between the original specification and the shipped product was large enough to kill 346 people in two crashes, less than five months apart.

For founders, normalization of deviance rarely ends in tragedy. It ends in the slow degradation of a standard you set when you were thinking clearly. The sales process that used to have four steps now has two because the team started skipping the ones that felt redundant. The quality check that used to happen before every release now happens monthly. The customer interview cadence that was supposed to be weekly is now "when we can get to it." Each cut was individually reasonable. The accumulated drift is not. The same dynamic plays out in team communication — when your team is filtering information because the brain treats dissent as a physical threat, each softened report normalizes the next, and the drift accelerates without anyone sounding the alarm.

The Creator's Blind Spot (Revisited)

The sunk cost problem is compounded by the same mechanism from Post 1: the more you've invested in building something, the less capable your brain is of evaluating it objectively.

The endowment effect inflates the value of what you own. The IKEA effect inflates the value of what you built. Sunk cost reasoning inflates the value of what you've already paid for. All three are running simultaneously in the founder who can't kill a failing product. The brain is stacking three separate biases in the same direction — keep going — and the conscious experience is a single feeling: this still has potential.

John Gourville's nine-to-one mismatch applies here too, but pointed inward. You're not just overvaluing what you've built relative to the market. You're overvaluing what you've built relative to what you could build next if you stopped. The opportunity cost of persisting — the thing you're not building because you're pouring resources into the thing that isn't working — is invisible to the same brain that can't let go of the sunk costs.

The Pivot That Paid

Stewart Butterfield knows what it feels like to kill something you built.

In 2012, Butterfield and his team at Tiny Speck had spent roughly three years and $17 million building Glitch, a browser-based massively multiplayer game. The concept was ambitious — a collaborative, nonviolent online world with a quirky aesthetic and an original creative vision. Butterfield had poured years of his life into it.

Glitch had a problem. It couldn't retain new users. The engagement curve wasn't a slope — it was a cliff. The team iterated, redesigned, relaunched. The cliff didn't move.

In November 2012, Butterfield shut Glitch down. He wrote a public letter to the community that was remarkably clear-eyed: "Unfortunately, Glitch has not attracted an audience large enough to sustain itself." Not "we're pivoting to a new direction." Not "we're exploring opportunities." The thing failed. He said so.

What happened next is the part every founder fixated on sunk costs should hear. During the years building Glitch, the team had developed an internal communication tool to coordinate across locations. It was fast, searchable, and organized around channels instead of email threads. After Glitch died, Butterfield looked at the tool and realized the internal product was more valuable than the one they'd been building for four years.

He pivoted. The internal tool became Slack. In 2021, Salesforce acquired Slack for $27.7 billion.

The $17 million and three years spent on Glitch weren't recovered. They were gone. The sunk cost was real. But the act of killing the failing product freed the resources — attention, engineering talent, creative energy — to build the thing that was actually working. The opportunity cost of persisting with Glitch was Slack. And the only way to capture that opportunity was to absorb the loss and let go.