The finding that made this famous is unpleasant and worth knowing. Patients undergoing a procedure without anaesthetic reported their discomfort minute by minute, and afterwards rated the whole thing. The rating tracked the worst moment and the final moment almost perfectly. It did not track the total, and it barely tracked the length — a procedure twice as long could be remembered as better, as long as it ended more gently.
That is not a rounding error in how people report things. It is what the memory is. And because the decision to come back is made by the memory rather than by the experience, the memory is the thing you are actually designing.
Below are two waits. Both of them are annoying, and one of them is meaningfully longer than the other. Sit through both, then say which one you would rather do again — and the page will show you what you actually sat through.
One person choosing once is not evidence
You made a single choice. Whatever it was, it proves nothing about you — you might have preferred the second one because it was second, or because you had worked out what the page was up to by then, which is a problem no version of this demo can design around. The original studies used hundreds of people who did not know what was being tested.
What the demo can honestly show you is the gap between the two totals, and how little that gap has to do with which one is easier to sit through. That part is not a claim about you. It is arithmetic.
What it costs you
Ending well is cheaper than being good throughout, and it is not the same thing. This is the most abusable finding in the whole collection. It can be read as licence to make a product that is bad in the middle and charming at the end, and that reading works, which is exactly why it deserves saying out loud rather than quietly using.
Find the peak before you polish the ending. Every flow has one moment that is worse than the rest — an error, a re-authentication, a form that lost its state. That single moment is doing more damage than the average of everything around it, and averages are how most teams look at their own funnels.
The end of the session is not the end of the experience. For most products the last moment is a receipt, a confirmation email, or a support reply three days later. Those are usually owned by whoever owns them, which is nobody.
Duration neglect cuts both ways. If length barely registers, the effort you spend shaving fifteen per cent off a flow is worth less than the effort spent on its worst moment. That is an unpopular thing to say in a room that has just built a performance dashboard.
I have shipped a cancellation flow that was faster and colder than the one before it. The numbers went the right way for a quarter. I don't think we ever measured the thing that mattered, and I'm not sure I'd know how.