Sunk cost fallacy
Money and effort already spent keep people going long after the case has gone.
People let unrecoverable past investment influence a forward-looking choice. Rationally, only future costs and benefits should matter, but having paid raises the felt cost of stopping because stopping makes the earlier spend feel wasted. The effect grows with how personally responsible the decider feels for the original commitment.
How it shows up in software
Onboarding that asks for real effort creates sunk cost in the user's favour and against it: someone eleven steps into a setup wizard will finish, and someone with three years of history in a tool will stay past the point where a competitor would serve them better. Internally it is worse. Roadmap items survive quarters because of the engineering already spent on them.
Using it well
- Show progress through multi-step setup so invested effort is visible and the remaining work looks finite.
- Save partial work by default and let users return to it, so past effort keeps its value instead of being an argument for pushing through now.
- Set kill criteria for a project before work starts, written down, so the shutdown decision is not made by the people with the most spent.
- In roadmap reviews, ask what you would fund if this work did not exist yet. Answer that question before looking at the burn to date.
Where it turns manipulative
- Long, effortful onboarding designed to trap rather than to configure. Making a user invest 20 minutes before showing them the price is a form of the roach motel pattern.
- Building switching cost by holding data hostage, such as export that produces a format nobody can import.
- Progress bars that overstate completion to keep someone in a flow they would otherwise leave.
Where you have seen it
LinkedIn
The profile strength indicator shows how complete a profile is and which sections remain.
TurboTax
The filing flow collects and saves a full return before the payment step for paid tiers.
Notion
Template galleries let a new workspace start populated, so early effort compounds into content rather than configuration.
What the research says
- Arkes and Blumer, 1985Well evidenced
Theatre subscribers randomly given a discount on their season tickets attended fewer plays over the season than those who paid full price, a field result on real money and real behaviour.
- Staw, 1976Mixed evidence
Participants who had personally chosen a failing business unit allocated more follow-on funding to it than those inheriting the same failing unit, which locates part of the effect in self-justification rather than accounting.
- Friedman et al., 2007Mixed evidence
Several controlled experiments failed to produce sunk cost effects in simple economic tasks, suggesting the effect depends on ambiguity about future returns and on personal responsibility.
The effect is real but conditional. It is strongest where the outcome is uncertain and the decider owns the original call, and it weakens or disappears in clean tasks with obvious payoffs. Animal and child studies give mixed results too.
Grades are a judgement about the evidence, not about how useful the idea is. Plenty of contested effects are still worth knowing, as long as you do not cite them as settled.