The Peak-End Rule: How Customer Memory Really Works

Most customer experience programmes measure the wrong thing. They track average satisfaction across a journey — averaging out the mundane, the frustrating, and the delightful into a single number that tells you almost nothing about what a customer will actually remember, or whether they will return. The psychology of memory doesn't work by averaging. It works by editing.

The peak-end rule, first described by the psychologist Daniel Kahneman and his colleagues in research published in the early 1990s, is one of the most consequential findings in behavioural science for anyone designing customer journeys. It states that people judge a past experience primarily by two moments: the emotional peak — the most intense point, positive or negative — and the ending. Everything in between is largely discarded by memory. Duration, on its own, barely registers.

The peak-end rule in one sentence: Customers do not remember the average of an experience — they remember how it felt at its most intense moment and how it ended, and those two data points drive every future decision they make about you.

This is not a peripheral curiosity. It is the operating system of customer memory. Understanding it changes how you design journeys, where you invest in service quality, and which moments actually earn loyalty versus which ones merely avoid complaints.

Where the science comes from

Kahneman and his colleagues demonstrated the peak-end rule through a series of studies involving genuinely unpleasant experiences — including a well-known cold-water experiment in which participants submerged a hand in painfully cold water for different durations and under different conditions. When asked which trial they preferred to repeat, participants consistently chose the longer one — provided it ended at a slightly warmer (though still uncomfortable) temperature. The trial that ended on a marginally better note was preferred, even though it involved more total discomfort by any objective measure.

This finding — that the ending overrides the duration in memory — has since been replicated across a wide range of contexts. Kahneman formalised the distinction between the experiencing self (who lives through an event moment by moment) and the remembering self (who constructs a narrative afterwards and makes future choices). In his TED talk on experience versus memory, Kahneman described this as the "tyranny of the remembering self" — it is the remembering self, not the experiencing self, that decides whether a customer comes back.

For CX practitioners, this distinction is not academic. Your Net Promoter Score, your CSAT survey, your post-interaction rating — all of these are measurements of the remembering self. They are snapshots of reconstructed memory, not real-time experience. Which means the peak and the end are disproportionately what you are actually measuring, whether you realise it or not.

Why averaging satisfaction scores is a design trap

Consider a bank customer who opens a new account. The process involves six touchpoints: an online application, a document upload, a branch visit, an identity verification call, a welcome kit delivery, and a first login to digital banking. Suppose five of those touchpoints are smooth and unremarkable. The branch visit, however, is a forty-minute wait followed by a terse interaction with a distracted adviser. And the final step — first login — is clean, fast, and ends with a personalised welcome message.

Average satisfaction across six touchpoints: reasonable. What the customer remembers: the wait at the branch (the negative peak) and the first login (the end). If the ending is strong enough, it can partially offset a painful peak. If the ending is poor, even a journey with five excellent touchpoints will be remembered as a bad experience.

This is why journey mapping that only tracks average scores misses the point. You need to know where the emotional peaks — positive and negative — actually occur, and what happens at the very end. Those are the moments that shape memory, word-of-mouth, and repurchase intent.

The negative peak problem: one bad moment rewrites the story

Loss aversion — the well-established finding from Kahneman and Tversky's prospect theory that losses feel roughly twice as powerful as equivalent gains — compounds the peak-end effect on the negative side. A single sharp moment of frustration, humiliation, or broken trust does not merely lower the average. It becomes the peak. And because negative peaks are weighted more heavily in memory than positive ones, a single service failure at the wrong moment can override multiple positive interactions.

This has a direct implication for customer crisis management: the moment of failure is not the only thing that matters. How the failure is resolved — and crucially, how the interaction ends — determines what the customer ultimately remembers. A complaint handled badly and closed abruptly creates a negative peak followed by a negative ending. A complaint handled with genuine empathy, resolved clearly, and closed with a meaningful gesture can convert a negative peak into a story the customer actually tells positively.

The ending is always within your control, even when the peak is not.

Designing for peaks: where to invest your service energy

If memory is shaped by peaks and endings, the logical design question is: which moments in your journey are most likely to become peaks, and are you engineering them intentionally?

There are two types of peaks worth designing for deliberately:

• Positive signature moments — interactions so unexpectedly good that they become the story a customer tells. These are not expensive to create; they are specific, personalised, and well-timed. A hotel that remembers a returning guest's coffee order. A government service that proactively notifies a citizen before a document expires. A retailer whose returns process is faster and friendlier than the original purchase. These moments work precisely because they violate the customer's expectation — which is the mechanism behind the affect heuristic: emotionally vivid moments are encoded more deeply.
• Negative peak mitigation — identifying the touchpoints most likely to generate frustration (long waits, opaque processes, unclear next steps) and either eliminating the friction or, where that is not possible, reframing the experience around it. A progress indicator during a long wait does not reduce the wait; it reduces the peak intensity of the frustration by giving the customer a sense of control.

The service design discipline has a useful tool for this: emotional journey mapping, which plots not just what happens at each touchpoint but how the customer feels. When you overlay an emotional arc on a journey map, the peaks — positive and negative — become visible. You can then make deliberate choices about where to concentrate design effort rather than spreading it uniformly across every touchpoint.

Designing for endings: the most underinvested moment in CX

Endings are systematically underinvested in most organisations. The commercial logic tends to front-load effort — acquisition, onboarding, the first transaction — and allow the tail of the journey to drift. But the peak-end rule makes this a costly error.

Consider which endings matter most in your category:

• The end of a service interaction (how does the agent close the call or conversation?)
• The end of a transaction (what does the confirmation screen or receipt communicate?)
• The end of a subscription or contract period (how is renewal or departure handled?)
• The end of a complaint resolution (what is the final message the customer receives?)
• The end of a physical visit (what is the last sensory impression as the customer leaves?)

Each of these is a memory-defining moment. A subscription cancellation flow that is gracious, fast, and leaves the door open will be remembered differently — and will generate different word-of-mouth — than one that is obstructive and cold. The goal-gradient heuristic suggests that people feel a sense of completion and satisfaction when they reach the end of a process cleanly; a journey that ends with ambiguity or friction leaves an open loop in memory, and open loops tend to be remembered negatively.

For teams working on customer loyalty, this is particularly relevant. Loyalty is not built by the average of every interaction — it is built by the accumulation of positive peaks and strong endings over time. A customer who has experienced three or four genuinely memorable positive moments, and whose interactions consistently end well, has a very different relationship with a brand than one whose journey has been uniformly adequate.

Applying the peak-end rule in practice

Translating this into operational CX design requires a few concrete shifts in how you work:

1. Audit your journey for emotional peaks. Use qualitative research — interviews, ethnographic observation, verbatim feedback — to identify which moments customers actually remember and describe. These are your real peaks, not the ones you assumed mattered. Voice of customer programmes that capture narrative, not just scores, are essential here.
2. Classify peaks as designed or accidental. Most negative peaks are accidental — nobody planned for the forty-minute wait or the confusing error message. Most positive peaks are also accidental — a particularly warm staff member, a lucky timing. The design goal is to make positive peaks intentional and repeatable, and to systematically eliminate the conditions that create negative ones.
3. Redesign your endings explicitly. For every major journey in your organisation, define what a good ending looks like. Script it, train for it, and measure it separately from the overall journey score. The closing moment of a service interaction deserves as much design attention as the opening.
4. Resist the pull of uniform improvement. When resources are limited, the peak-end rule tells you where to concentrate. A marginal improvement to a mid-journey touchpoint that is never remembered will have less impact on customer memory and loyalty than a significant improvement to the peak or the ending.
5. Use emotional arc mapping as a standard tool. Plot customer emotion — not just satisfaction — across the journey. The shape of that arc, and where it peaks, is the diagnostic that tells you where memory is being formed.

The ethical dimension: peaks can be engineered for harm as well as good

It would be incomplete to discuss the peak-end rule without acknowledging its darker applications. The same mechanism that allows a brand to create a genuinely delightful memory can be used to manufacture artificial peaks — a dramatic countdown, a manufactured scarcity moment, a manipulative upsell — that distort customer decision-making without delivering real value.

Behavioural science applied to CX design carries an ethical obligation: the goal is to design experiences that customers remember accurately and positively because the experience genuinely merited it — not to exploit memory distortion to override rational evaluation. The distinction between a nudge and a manipulation is whether the intervention serves the customer's interests or merely the organisation's. Designing a strong ending because it genuinely closes a journey well is good design. Engineering a false emotional high to obscure a poor product is not.

The most durable competitive advantage from the peak-end rule comes from the former. Customers who remember an experience accurately and warmly are the ones who return, refer, and remain loyal through competitive pressure. Manufactured peaks tend to be fragile — they collapse the moment the customer reflects more carefully, or compares notes with someone else.

Memory, ultimately, is the currency of loyalty. Every journey you design is writing a story in the customer's mind — and the peak-end rule tells you that only two sentences of that story will be reliably recalled. The question worth asking is not "how was the average?" but "what did we make them feel at the sharpest moment, and how did we say goodbye?" Get those two things right, and the rest of the journey has room to breathe.