A support ticket closes. The refund clears. The agent types "Glad I could help." Every metric in the system registers a success: first-contact resolution, sub-four-minute handle time, a 9 on the post-chat survey.
Four months later the account doesn't renew. Nobody can explain it, because nothing in the record says anything went wrong.
Nothing did go wrong. The problem is that the customer who filled out that survey and the customer who declined to renew were, in a meaningful sense, two different people. One lived the experience. The other remembers it. Only the second one has a wallet.
Daniel Kahneman spent much of his later career on a distinction most marketing organizations have never absorbed: the experiencing self and the remembering self. The experiencing self exists only in the present. It registers each moment as it happens and then it is gone. The remembering self is the one that keeps score, tells the story, and makes the next decision.
These two selves do not agree. In a randomized trial published in Pain, Redelmeier and Kahneman had patients rate discomfort in real time during colonoscopies, then rate the whole procedure afterward. The retrospective judgment tracked almost perfectly with two numbers: the worst moment and the final moment. It tracked with total duration at close to zero. Patients whose procedures lasted three times as long did not remember them as three times worse. They did not remember them as worse at all.
The finding has a name — the peak-end rule — and a companion, duration neglect. It is not a curiosity confined to gastroenterology. A 2022 meta-analysis in Organizational Behavior and Human Decision Processes pooled 174 effect sizes and found the peak-end effect both large (r = .58) and robust across every boundary condition the authors tested — while the effect of duration was, in their words, essentially nil. The peak and the ending dominate. The middle evaporates.
Which means the average is a fiction. Not a rough approximation, not a lossy summary — a number describing an experience no human being ever had.
Now look at how the average firm measures customer experience. CSAT at ticket close. Session-level satisfaction. Journey analytics that weight every stage by volume. NPS at a fixed interval, usually chosen for reporting convenience rather than for psychological relevance.
Every one of those instruments samples the experiencing self. Then the business uses the result to forecast the behavior of the remembering self. That is a category error, and it is expensive.
The long argument over NPS has mostly been an argument about arithmetic — whether an eleven-point scale is defensible, whether the promoter/detractor cut points are arbitrary, whether the original growth correlations survive replication. They largely didn't; careful reviews of the literature put NPS roughly in the same modest predictive range as ordinary satisfaction measures, well short of the claims made for it. Fine. But the scale is the least interesting flaw.
The interesting flaw is when you ask. A survey fired at ticket close captures a customer who has just been relieved of a problem. That is a local maximum, and it is not the moment that will govern their memory six weeks later. You are measuring at the point of maximum institutional convenience and minimum predictive value, then treating the output as a leading indicator.
Meanwhile the outcomes keep drifting the wrong direction. Forrester's 2025 Global Customer Experience Index, built on more than 275,000 customers across 469 brands, found 21% of brands declining and only 6% improving. In the US the split was worse — a quarter down, seven percent up, and quality eroding across effectiveness, ease, and emotion simultaneously. Forrester's own framing is that the gap between the experience brands intend to deliver and the one customers actually register is widening.
That gap is not a delivery failure. It is a measurement failure that produces a delivery failure. If your instruments report an average, your investment will follow the average, and you will spend evenly across a journey that memory records unevenly.
Duration neglect has a blunt budget implication that almost nobody acts on: extending a good experience buys you very little. Intensifying one buys you a great deal.
Most experience budgets are allocated the opposite way — spread thin across every touchpoint, because every touchpoint appears in the journey map and every touchpoint has an owner who wants funding. The result is a journey with no peak. Competent throughout, memorable nowhere. A customer who cannot recall a single specific moment of working with you has no material from which to construct a preference, and will therefore make the renewal decision on price, because price is the only variable they can actually retrieve.
The counterexample is unglamorous. Atour, a Chinese hotel chain, rebuilt its experience strategy explicitly around the peak-end rule. Documented in a 2022 case study, the redesign concentrated on check-in and check-out — personalized greeting, a scented lobby, a genuine send-off at departure — and produced higher satisfaction, more repeat stays, and better reviews. As Mark Levy noted in CX Dive, they didn't add pillows or upgrade the Wi-Fi. They fixed the two moments memory actually keeps.
Note what that costs relative to a full-journey improvement program. Note also how few competitors will copy it, because it doesn't look like a strategy. It looks like a nice touch.
Here is where most brands quietly forfeit the argument.
The last moment of a customer relationship is not the last invoice. It is the cancellation. It is the offboarding email, the data export request, the retention flow that makes leaving harder than joining. That is the final data point the remembering self files — and it is the one that gets loaded into every review, every referral conversation, every reply when a peer asks whether you're any good.
The industry just ran a live experiment on this. On July 8, 2025, the Eighth Circuit vacated the FTC's "click-to-cancel" rule on procedural grounds, days before compliance was due. The FTC has since restarted the rulemaking, filing an advance notice in January 2026, and ROSCA plus state auto-renewal statutes never stopped applying. But for a window, the specific federal mandate simply vanished.
Most companies that had built symmetrical cancellation flows in anticipation had a decision to make. A meaningful share rolled the friction back. That decision was filed under compliance, and it was never a compliance question. Cancellation friction is a wager: that the revenue extracted from customers too tired to keep clicking exceeds the cost of every one of them narrating that experience for the next decade.
It is a bad wager under peak-end logic, and the reason is arithmetic. The retained revenue is bounded and near-term. The memory is unbounded and permanent. You are trading a quarter of churn deferral for the final impression that will represent your brand in every conversation that customer ever has about your category. And they will have those conversations at exactly the moments a prospect is deciding.
An exit designed with dignity — one screen, no interrogation, data exported without a support ticket, a straight "we'd take you back" — is the cheapest brand asset available to most companies. It costs an engineering sprint. It buys a permanent shift in the story.
Move the survey off the ticket and onto the memory. Stop asking at resolution. Ask at a delay — far enough out that the remembering self has consolidated the story, close enough that it hasn't decayed. Then ask retrospective questions rather than transactional ones: what stands out, how it ended, whether they would describe it to a colleague. You are trying to read the record that will drive behavior, not the receipt.
Find your peak and fund it disproportionately. Audit the journey for the single highest-intensity moment, positive or negative. If the peak is negative — the outage, the billing dispute, the implementation week — that is where the entire budget belongs, because a negative peak cannot be offset by competence elsewhere. If there is no peak at all, you have a bigger problem than any of your dashboards will report: you are forgettable, and forgettable is priced like a commodity because it is one.
Treat the exit as a campaign asset. Give offboarding an owner, a design review, and a budget line. Measure it. The last impression is the only one that compounds, and it is currently being designed by a retention team optimizing against a metric that ends the day the customer leaves.
None of this is a call for theatrics. Engineered delight that a customer can see being engineered doesn't register as a peak; it registers as a tactic, and it gets filed accordingly. The moments that hold are the ones that appear to cost something real — attention, restraint, a decision made in the customer's favor when the company had every incentive to decide otherwise. A frictionless cancellation reads as costly precisely because everyone knows the alternative was more profitable in the short run.
That is the underlying logic. Memory doesn't record effort. It records intensity and finality, and it treats both as evidence of what you are actually like when nobody is auditing.
The dashboard will keep reporting an average. It will keep being wrong in the same direction, quarter after quarter, and the variance will keep getting explained as market conditions. The customer who renews was never in that data. They were reconstructed after the fact, out of two moments you probably didn't choose, and they made their decision on the basis of that reconstruction alone.
You can pick those two moments. Almost nobody does.
InPhluence builds campaigns and communications strategy on behavioral evidence rather than convention. If your experience data says one thing and your renewal rate says another, the gap is usually structural. Start a conversation.