Design the ending, not the average

Here is one of the more uncomfortable findings in decision science, from a study you would not volunteer for. In the era when colonoscopies were done without much sedation and genuinely hurt, researchers tracked patients’ moment-to-moment discomfort and then asked, afterward, how bad the whole thing had been. Two patients make the point. One had a short procedure that ended abruptly at a moment of sharp pain. The other had a longer procedure — more total discomfort, by every objective measure — but the final few minutes were deliberately mild. The second patient remembered the experience as less awful, and was more willing to come back. More pain, better memory. Because memory doesn’t add up the minutes or average them. It takes two snapshots: the worst (or best) moment, and the end.

Those are two findings, and both matter. The peak–end rule: your memory of an episode is approximately its most intense point plus its final moment. And duration neglect: how long it went on barely registers in the remembered version. Put them together and what you lived and what you remember come apart — and the version that drives your future decisions, your willingness to return, the story you tell other people, is the remembered one, built from just those two points.

The sharpest application, and the most neglected, is the ending — and the ending nobody designs is the exit. Teams pour craft into onboarding, the beginning, the first ten minutes, and then treat the cancel flow as an afterthought or, worse, a hostage negotiation: buried buttons, guilt-trip copy, a retention gauntlet you have to fight through to leave. But a cancellation is an end, a high-salience, memory-forming moment, and it’s the note the person carries into every future conversation about you. Make it graceful — here’s your data, the door’s unlocked, come back whenever — and they remember the whole relationship warmly, and sometimes return, and speak well of you meanwhile. Make it cruel and you don’t just lose them; you poison the remembered average of everything that came before, and convert a quiet churned user into an active anti-referral who tells the cancel story at dinner. You are not designing to trap the person leaving. You’re designing the memory they’ll narrate about you for years.

The peaks deserve the same intentionality. Make sure there is a genuine high point — the moment the product’s value lands, the “aha” — because a flat, uniformly pleasant experience with no peak gives memory nothing bright to grab. And watch that your sessions don’t accidentally end on their worst moment: a failed export, a sudden paywall, a dead end. If the last thing a user touches before they close the tab is friction, that friction is what tomorrow’s memory is built from. End on competence — a saved file, a finished task, a small win — not on a wall. Duration neglect, meanwhile, cuts kindly in both directions: a long, boring necessary stretch is remembered less harshly than you fear, so don’t panic-optimize every second of it; and a long good stretch doesn’t bank memory-credit in proportion to its length, so a strong finish earns you more than simply adding more fine minutes.

I have to flag the real caveat hard, because this idea is a short step from something cynical. This is a rule about remembered experience, and memory is not everything. People also live in the moment-to-moment stream, not only in the highlight reel of it, and a product that’s quietly miserable to use every day but ends each session on a confetti burst is optimizing the wrong thing — users feel that gap eventually, and the staged warmth starts to read as exactly the manipulation it is. Use peak–end to stop neglecting your endings, not to paper over a rotten middle. A genuinely good experience with a well-designed peak and a gracious end is the goal; a bad one wearing a fake smile at the door is a trick with a short shelf life. And be honest that the science, clean as it is for short painful episodes, gets murkier for long, layered relationships with software.

So the next time you’re deciding where to spend a week of polish, look past the screens users spend the most time in — memory barely counts that — and toward the two moments they’ll actually keep: the highest point, and the last one. Especially the last one. Including the goodbye you almost certainly haven’t designed at all.