The ending has the highest return on investment

What it is

Under a fixed experience budget — the same staff time, effort, and surprise resources — spending it on the final stretch of an experience returns more than spending it in the middle. This is not a general reminder to "care about the wrap-up"; it is a placement rule derived from the peak-end rule: retrospective weight concentrates on the peak and the end, mid-course spending is nearly erased at reconstruction, and ending spending enters memory at close to full weight. Where the budget goes decides the exchange rate at which it converts into memory.

Why it happens

The mechanism is the asymmetry between two exchange rates. An improvement placed mid-course raises one stretch of the real-time curve, and the middle is exactly what snapshot storage discards — duration neglect spreads evenly distributed improvements thin, toward zero in the memory account. An improvement placed at the end rewrites a heavily weighted item of the memory representation and enters the retrospective score nearly one to one. There is a further layer: the ending is the last frame the user takes away and the natural landing point of retelling — "they even brought dessert," "the cancellation page was refreshingly clean" — and retelling amplifies the ending further, with no mid-course equivalent. Same budget, better rate at the end, so the optimal placement presses discretionary resources toward the close.

Studying it

• Ending-manipulation experiments: hold the first part of an experience stream identical and change only the ending (for instance, appending a slightly milder tail), then compare retrospective ratings and behavioral choices between groups. In the classic results, the group given the better ending not only rated the episode higher but also systematically chose to repeat the longer version — more total discomfort, better end — over the shorter one; both ratings and choices follow the ending.
• Field follow-up version: adding a brief, gentled finish at the end of a real medical procedure shifts the intervention group's remembered evaluation at follow-up, and behavior — willingness to come back — moves with it.
• Methodological cautions: judge ending manipulations on behavioral outcomes, not only on ratings — a small rating difference can amplify into a large gap in binary choices like returning or repurchasing; conversely, an unmoved immediate rating in a field experiment is not failure, because behavior lags attitude — allow an adequate follow-up window.

Where it stops holding

• Manipulable does not mean fabricable. A perfect finish disconnected from the rest of the experience, or recognizable as a script, reads as a bribe: the detection itself becomes a new negative moment, and the retrospective evaluation drops instead of rising. Ending investment presupposes proportionality to what the experience actually was — amplifying existing good, not patching on a flawless facade.
• A good ending does not compensate for harm in the process. A towering negative peak (data loss, humiliation, a severe outage) is not wiped out by a warm ending; the rule specifies a weighting structure, not "redemption at the last moment." Ending improvements rewrite the evaluation only when commensurate with the peak's magnitude.
• In long-lived products "the ending" is plural. The close of each session, the task completion state, the order's final page, even cancellation and unsubscribe flows are each endings. Grooming only the signup funnel's finish while ignoring everyday session closes invests one endpoint and skips the rest.
• Evidence boundary. The conclusions come from short experiences with fully controlled stimulus streams; in real services users bring their own goals and expectations, and both the controllability of the ending and the effect size shrink — verify on-site rather than importing laboratory magnitudes.

Applying it

• Inventory every touchpoint of the journey and shift discretionary delight budget to a tail-heavy allocation: confirmation and result pages, the completion moment, the last step's copy and give-aways outrank any mid-course decoration.
• Design an explicit closing moment for each core flow: do not stop dead on the final click — provide the result, where to go next, and a next step, so the user leaves in a state of completion; that frame is the one they carry away.
• Spend on exit paths too: the last screen of cancel, unsubscribe, and sign-out flows decides how users retell their leaving; reserve budget for them.
• Tail-heavy spending replaces mid-course polish, not mid-course repair: remove the valleys first, then press resources toward the close.
• To validate: run an equal-budget controlled comparison — one version spreads the budget evenly, one is tail-heavy — and compare retrospective ratings plus return and recommendation intent; add a one-week-later follow-up where possible, since remembered evaluations diverge further from in-the-moment ones than an immediate survey shows.

Related

• Same group: P1.03.1 Overall evaluation is dominated by the peak and the end · P1.03.3 Negative peaks are amplified in memory too
• Nearby: P1.09.4 Recovery after failure is a low-cost peak position · P1.09.1 Delight intensity must be proportionate to the event's actual weight · P1.02.3 Delighters must not extend the critical path
• Search terms: peak-end rule · last impression · ending on a high note · completion state · experience budget