Peak-end rule
People judge a past experience by its most intense moment and its final moment.
When people rate an experience from memory, the average of the peak intensity and the intensity at the end predicts the rating better than the total amount of pleasure or pain. How long the experience lasted barely moves the rating, an effect called duration neglect. The rule describes remembered experience, not experience as it was lived.
Two trials, same peak, different ending
How it shows up in software
Product teams reach for this when designing checkout, cancellation, support tickets and error recovery. The claim is that a flow with one bad moment and a clean ending will be remembered better than an evenly mediocre one. Extending it to multi-day onboarding or a year of subscription use is extrapolation from minute-long aversive episodes, not a finding.
Using it well
- Find the single worst moment in a flow by watching session recordings or reading support tickets, then fix that moment before smoothing anything else.
- Give long or painful flows a real ending: a confirmation screen that states what happened, what the person now has, and what happens next.
- When a task fails, end on recovery. Preserve what they typed, say what went wrong in their words, and put the retry action under the cursor.
- Measure the lived experience too. Track error counts and time on task alongside satisfaction scores, since the rule predicts the rating, not the reality.
Where it turns manipulative
- Stacking a delightful ending onto a flow you know is slow and error-prone. You are buying a better score while the underlying experience stays broken, and the fix costs less than the confetti.
- Engineering the ending of a cancellation flow to feel warm so the person forgets how hard cancelling was. The friction they endured is the thing to remove.
- Treating a good final screen as licence to let the middle of a flow degrade, on the theory that nobody remembers the middle. Support volume remembers the middle.
Where you have seen it
Duolingo
A lesson closes on a summary screen with the streak, the XP earned and an animation, so the session ends on a resolved beat rather than on the last question answered.
Stripe Checkout
A declined card keeps the entered details in place and shows the reason inline, so the flow ends on a correctable state instead of a cleared form.
Apple Music
Cancelling a subscription ends on a screen stating the exact date access continues to, so the last moment is information rather than a plea.
What the research says
- Kahneman, Fredrickson, Schreiber and Redelmeier, 1993Well evidenced
Participants held a hand in painfully cold water for 60 seconds, and in a second trial for 60 seconds plus 30 extra seconds as the water warmed slightly. Most chose to repeat the longer trial, which contained more total pain but a milder ending.
The stimulus is a laboratory cold-pressor task lasting about a minute. It shows how a short aversive episode is summarised, not how someone remembers a week of using software.
- Redelmeier and Kahneman, 1996Well evidenced
Patients logged discomfort minute by minute during colonoscopy. Retrospective ratings tracked the peak and final minutes; total procedure duration, which ranged from four to 69 minutes, had little effect.
- Redelmeier, Katz and Kahneman, 2003Mixed evidence
A randomised trial added a few minutes of low discomfort to the end of colonoscopies. The extended group remembered the procedure as less unpleasant and returned for follow-up at a higher rate.
One clinical trial in one procedure. Treat the return-rate result as suggestive.
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.