Sunk Cost Fallacy
Date: 2026-09-27
Continuing with something because of what’s already been put into it, rather than what it’ll return from here. Customers do it with subscriptions, part-filled forms and loyalty points; teams do it with failing projects and losing tests. Only the second is yours to fix.
A sunk cost is a cost already paid that can’t be recovered whatever you decide next. The sunk cost fallacy — also called the sunk cost effect — is letting that cost influence the decision anyway: continuing because of what’s been invested rather than what’s to come.
Origin: named as a behavioural effect by Hal Arkes and Catherine Blumer, “The Psychology of Sunk Cost” (1985). The economist’s rule it violates — ignore sunk costs — is much older; see Marginal Analysis.
The theatre study
Arkes and Blumer sold season tickets to a university theatre at three prices: full price, and two discounts, assigned at random.
FULL PRICE DISCOUNTED
plays, identical same same
tickets, identical same same
attended, first half ~4.1 plays ~3.3 plays ← paid more, went more
of the season
Randomisation means the groups didn’t differ in how much they liked theatre — only in how much they’d sunk. The difference faded in the second half of the season. [CHECK: exact figures and the second-half result against the paper.]
What it’s evidence for: the money was gone either way, so attending more to “get your money’s worth” is the fallacy in the wild. And the effect fades — sunk costs lose their grip with time.
Evidence status: the effect is well-replicated in lab vignettes and some field data, but a recent study found individual sunk-cost vignette responses unreliable across repeated measures — robust as a population tendency, weak as a trait of a person. [CHECK: current state of the sunk cost replication literature.]
Where customers do it
| Pattern | What’s sunk | Honest use | Manipulative use |
|---|---|---|---|
| Multi-step forms | Effort already given | Easy steps first, honest progress | Hiding the cost until step 4 — Commitment and Consistency |
| Loyalty points | Past spend | Clear value, reachable rewards | Expiry designed to force a purchase — Loyalty Programmes |
| Subscriptions | Months paid; setup effort | Saved preferences make staying easier | ”You’ll lose everything” on cancel — Cancellation Flows |
| Configured products | Time configuring | Save the configuration | Timed-out configurations |
| Annual plans | Up-front payment | Better price for commitment | — |
It’s close kin to loss aversion — abandoning means realising the loss — and customers often experience it as consistency (“I’ve come this far”). The three explanations overlap and a design rarely engages only one — Loss Aversion.
Where teams do it
The version that actually costs money inside a business:
- Running a losing test longer “to give it a chance” — the traffic already spent isn’t a reason to spend more. Decide stopping rules before starting — Stopping Rules
- Shipping a feature because it was built — the build cost is gone; the question is whether it earns its maintenance from here — Technical Debt
- Continuing a replatform or agency relationship past the point the evidence says stop — Replatforming
- Defending a campaign on what it cost rather than what it returns at the margin — Marginal Analysis
The diagnostic question: if we were starting from here today, with nothing invested, would we choose this? If not, the investment is the only thing keeping it alive.
Where it goes wrong
- Calling everything sunk cost. Sometimes continuing is right because the remaining cost is small and the remaining value is large — that’s marginal thinking, not a fallacy
- Ignoring real switching costs. Retraining, migration and data loss are future costs, and they legitimately count. Only the unrecoverable past is sunk
- Designing retention around it. Customers kept by what they’ve sunk rather than what they get leave loudly once they notice — Retention and Churn