Peak-end rule = people judge a past experience by its most intense moment and by how it ended, not by its average.
The problem: it has turned into a licence to skip the boring parts. Find the peak, polish the ending, let the middle rot. That is not what the studies say.
Memory compresses your product. It does not refund the parts it left out.
1. Length really does count for almost nothing
What teams ship: shorter flows, fewer steps, measured in seconds saved.
Why it fails: a 2022 meta-analysis pooled 174 effect sizes. The peak-end effect on how people rated a past experience was large, r = 0.581. The effect of how long the experience lasted was close to zero.
Ship instead: stop selling "we cut six steps to four" as an experience win. It is an efficiency win. Say so.
2. The middle is not free
What teams ship: one hero moment, one confetti screen, and a middle nobody owns.
Why it fails: in that same meta-analysis, the peak-end effect was about the same size as the effect of the plain average of the whole experience. A 2024 review of mental health studies found several where the average predicted what people remembered better than peak, end, or the two together.
Ship instead: treat peak and end as the cheapest places to buy memory, not the only ones. A rotten middle still lands in the score.
3. Your ending is not the success screen
What teams ship: the last frame anyone designed is the confirmation page. After that the user is on their own.
Why it fails: the real ending is the wait, the email that turns up later, the cancel flow, the empty state on day two. In exposure therapy, closing a session on a milder version of the feared thing improved how people later rated the distress of the whole session. The ending is a design surface.
Ship instead: write down where your experience actually ends. Then give that screen the care you gave onboarding.
4. Peak and end do not simply add up
What teams ship: "improve the peak" and "improve the ending" as two separate tickets.
Why it fails: a 2026 study on a word-learning app tested it. How people felt about the task tracked the peak-end average far more closely (r = 0.761) than the average of their page-by-page ratings (r = 0.314). But when the team improved the peak page and the end page together, the two changes did not stack the way two tickets assume. The first study ran on 32 people, so hold it loosely.
Ship instead: test the pair, not the parts.
5. There is a line, and it is nearer than you think
What teams ship: a padded ending, because a gentler last thirty seconds lifts the survey score.
Why it fails: a 2025 ethics paper asked whether a doctor may stretch out a painful procedure purely to leave a better memory of it. It could not settle the question. Benefit pulls one way, consent pulls the other. Your version is making someone's experience longer, or worse in the moment, so it scores better afterwards.
Ship instead: change the ending by taking pain out, never by putting time in.
What the room says, and what the evidence says
| Said in the deck | What the research supports |
|---|---|
| "People only remember the peak and the end" | Peak and end weigh heavily. So does the plain average. |
| "Length does not matter" | Closest to true. It barely moves the memory. |
| "Fix the peak, fix the ending, done" | The two interact. Test them together. |
| "It always applies" | Across 1,889 people in daily life, peak and end effects showed up in about half the conditions tested. |
| "A better ending is a free win" | Not if you bought it by making the experience longer. |
The ten minute audit
Open your product. Write down the single most intense moment a user has this week, and the last thing that happens to them. If you cannot name both, that is the finding. If you can, ask who owns them. In most teams the answer is nobody.
Get the PDF