The Peak-End Rule in Customer Experience

How people remember an experience is not the same as how they experienced it. This gap between the experiencing self and the remembering self has profound implications for product design.

Anika van der Berg · August 5, 2025

Two Selves

In a series of studies in the 1990s, Daniel Kahneman and colleagues demonstrated a striking dissociation between experienced utility — how people feel moment-to-moment during an experience — and remembered utility — how people evaluate the experience in retrospect. The most famous of these studies involved a colonoscopy procedure. Redelmeier and Kahneman (1996, N = 287) found that patients' retrospective evaluations of pain were predicted almost entirely by two factors: the peak intensity of pain during the procedure and the intensity of pain at the very end. The total duration of the procedure — whether it lasted 8 minutes or 69 minutes — had essentially no effect on remembered pain.¹

This is the peak-end rule: retrospective evaluations of experiences are dominated by the most intense moment (the peak) and the final moment (the end), with duration largely neglected. It has been replicated across aversive experiences (cold water immersion, loud noises, medical procedures) and positive experiences (vacations, meals, entertainment).

The implication for customer experience is fundamental. When users evaluate a product — when they decide whether to renew, recommend, or review it — they are consulting their remembering self, not their experiencing self. They are not aggregating every moment of interaction. They are recalling the high point (or low point) and the most recent interaction. Everything else fades into the background.

Peak-End in SaaS Onboarding

The onboarding experience is, for most SaaS products, the period that determines whether a user becomes a customer or a churn statistic. It is also the period where peak-end dynamics are most consequential.

Consider a typical 14-day trial. The user encounters dozens of interactions: sign-up forms, welcome emails, feature tutorials, integration setups, support conversations, and eventually, the conversion prompt. The peak-end rule predicts that the user's overall evaluation of the trial — the judgment that determines whether they convert — will be disproportionately influenced by the best (or worst) single moment and the final interaction before the conversion decision.

A 2013 study by Dixon, Toman, and DeLisi at the Corporate Executive Board (now Gartner) analyzed customer effort data from over 97,000 customer interactions across multiple B2B companies. They found that the strongest predictor of customer loyalty was not the average quality of interactions but the absence of high-effort, frustrating interactions — the negative peaks.² A single terrible support experience, a single confusing workflow, a single moment of "I cannot figure out how to do this" had more impact on loyalty than dozens of smooth, unremarkable interactions.

This finding aligns with the peak-end rule's emphasis on extremes. The implication is that reducing negative peaks may be more important for retention than increasing the average quality of the experience. A product that is consistently adequate but never terrible may generate better retrospective evaluations than a product that is sometimes excellent but occasionally awful.

Engineering the Peak

If retrospective evaluation is dominated by the peak moment, then deliberately designing a peak moment into the customer experience becomes a strategic priority. In the customer experience literature, this is sometimes called the "signature moment" or "defining moment" — a single interaction that is markedly better than the baseline and that comes to represent the brand in the customer's memory.

Chip and Dan Heath, in their book "The Power of Moments" (2017), argue that organizations should invest disproportionately in creating peak moments rather than spreading resources evenly across the experience. Their analysis draws on research by Ariely and Carmon (2000), who showed that the timing and intensity of peaks matter more than the cumulative quality of an experience.

In the SaaS context, the "aha moment" — the point at which a user first experiences the core value of the product — is the natural peak candidate. For Slack, this might be the first time a team conversation replaces an email thread. For Figma, it might be the first real-time collaborative editing session. For a data analytics tool, it might be the moment a complex query returns an insight that would have taken hours to produce manually.

Research on the aha moment in SaaS is largely proprietary, but some data points have been shared publicly. Chamath Palihapitiya, former head of growth at Facebook, famously identified "7 friends in 10 days" as Facebook's aha moment — the point at which users had enough social connections for the product to become sticky. Dropbox identified file synchronization across two devices as their critical activation event. In each case, the aha moment represents the peak of experienced value during onboarding, and reaching it is the strongest predictor of retention.

The peak-end rule adds a layer to this analysis: it is not enough for the aha moment to occur. It needs to be emotionally salient — a moment that rises above the baseline experience and registers in memory as a distinct positive peak. A quiet realization that the product is useful is less powerful than a moment of genuine delight or surprise. This is why the best onboarding experiences do not just deliver value but dramatize it: showing users exactly what they accomplished, quantifying the time saved, celebrating the milestone.

Engineering the End

The end of an experience has a disproportionate effect on retrospective evaluation, which creates both an opportunity and a risk for SaaS products.

The opportunity: the final interaction before a user evaluates the product can be deliberately designed to leave a positive impression. For trial conversions, this means that the last touchpoint before the conversion prompt — whether it is a summary of value delivered during the trial, a personalized recommendation, or simply a well-designed conversion page — has outsized influence on the decision.

I was involved in a redesign of trial-end communications for a marketing automation SaaS in 2023. The original approach was a standard "your trial is expiring" email with a payment link. We replaced it with a personalized summary: "During your 14-day trial, you created 7 campaigns, sent 12,400 emails, and saw a 34% open rate." The email then asked: "Would you like to keep these campaigns running?" This redesign increased trial-to-paid conversion by 17% (N = 3,200 trial users, split test over 8 weeks). The mechanism, I believe, was not merely informational — it was a peak-end intervention that reminded the user of their peak value moments right at the endpoint of the trial experience.

The risk: poor endings disproportionately damage the overall evaluation. In SaaS, the most common poor endings are involuntary — the trial expires abruptly, data is deleted without warning, features are locked mid-workflow. These negative endings can undo the goodwill built during an otherwise positive trial. Kahneman's colonoscopy research showed that a procedure ending with a period of reduced pain (even if it extended the total duration) was remembered more favorably than a shorter procedure ending at peak pain. The analogous principle for SaaS is that a graceful trial ending — gradual feature reduction rather than abrupt lockout, data preservation rather than deletion — will be remembered more favorably even if the user does not convert immediately.

Duration Neglect and Its Consequences

The flip side of the peak-end rule is duration neglect: the observation that the total duration of an experience has minimal impact on its retrospective evaluation. In Kahneman's cold water studies (Kahneman, Fredrickson, Schreiber, & Redelmeier, 1993, N = 32), participants preferred a 90-second trial that ended with a period of slightly warmer water over a 60-second trial that ended at the coldest temperature, even though the longer trial contained strictly more total discomfort.³

For SaaS customer experience, duration neglect suggests several counterintuitive conclusions. A long support interaction that ends with a clear resolution may be remembered more favorably than a short interaction that ends with ambiguity. A lengthy onboarding process that culminates in a significant aha moment may be preferred to a quick onboarding that never delivers a peak. Time spent in the product is less important than the intensity and valence of the most memorable moments in the product.

This does not mean that duration is irrelevant to the experiencing self — long wait times, slow load times, and tedious processes are genuinely unpleasant in the moment. But they may matter less to the remembering self than product teams assume, provided the peak and end are well-designed.

Caveats and Limitations

The peak-end rule was discovered in the context of temporally bounded, continuous experiences — medical procedures, cold water immersion, short films. A SaaS product relationship is not a continuous experience; it is a series of discrete interactions distributed over weeks, months, or years. Whether the peak-end rule applies in the same way to extended, discontinuous experiences is an open empirical question.

There is some evidence that it does. Research on vacation satisfaction by Wirtz, Kruger, Scollon, and Diener (2003) found that retrospective evaluations of vacations were predicted by peak moments and end moments, consistent with the peak-end rule, even though vacations are multi-day, multi-activity experiences. But a SaaS subscription is even more extended and fragmented than a vacation, and the generalization should be treated as a hypothesis, not a certainty.

Additionally, my discussion has focused primarily on positive peaks and positive endings. The interaction between negative peaks, positive endings, and overall evaluation is complex and context-dependent. A product that delivers a terrible experience on day 3 but recovers by day 14 may or may not benefit from a peak-end recency effect. The literature is not definitive on this point.