The peak-end rule

In short

When you remember an experience, your mind doesn’t take the mean of all its moments. It holds on to two moments: the most intense and the last. This phenomenon is called the peak-end rule.

According to the rule, the memory lies close to the mean of the worst and the last moment. How long the experience lasted takes up little room in memory. This is called duration neglect.

For product design, the meaning is clear: people make their next decision based on their memory. In an order flow, two places shape this memory most: the worst screen and the last one.

The cold-water experiment (1993)

One of the experiments supporting the rule appeared in 1993 in the journal Psychological Science. The authors are Daniel Kahneman, Barbara L. Fredrickson, Charles A. Schreiber and Donald A. Redelmeier. The paper is titled “When More Pain Is Preferred to Less: Adding a Better End”.

32 male university students took part. Each participant put a hand in cold water twice. In the short trial, the hand stayed in water at 14 °C (about 57 °F) for 60 seconds.

In the long trial, the other hand also stayed in water at 14 °C for 60 seconds. Then it stayed in the water 30 seconds longer, while the water was slowly warmed to about 15 °C (59 °F). It still hurt, but distinctly less for most participants.

So the long trial contained all the pain of the short one and added 30 seconds. Participants were then asked which trial they wanted to repeat. 22 of 32, or 69%, chose the long trial.

The choice was linked to how much the discomfort dropped in the last 30 seconds. Of the 21 participants whose discomfort fell by at least 2 points, 17 chose the long trial. In a replication without continuous measurement, 24 of 37 people made the same choice.

The authors’ explanation: the mean of the worst and the last moment predicts with fair accuracy how the experience is later rated. Duration, by contrast, is relatively neglected. Most participants knew correctly which trial had lasted longer, but didn’t use that in their choice.

The colonoscopy study (1996)

The second study is the paper by Donald A. Redelmeier and Daniel Kahneman that appeared in 1996 in the journal Pain. Pain was recorded continuously during the procedure for 154 colonoscopy patients and 133 lithotripsy patients (breaking up kidney stones).

The patients then rated the total pain. How much pain was remembered varied widely from person to person. The rating was linked to the most intense pain of the procedure and to the pain in the last 3 minutes.

According to the paper’s abstract, long procedures weren’t remembered as particularly unpleasant. The 1993 paper also summarises a colonoscopy study by the same two authors, which had been submitted for publication at the time. There, procedures lasted between 4 and 69 minutes, and duration didn’t significantly affect the later rating.

I report these two studies only to show how memory works. This page doesn’t give medical information or advice.

How to calculate the peak-end estimate

The 1993 paper also summarises a 1992 study by Varey and Kahneman. There, participants rated sequences of discomfort ratings from 0 to 10. They had been told the sequences came from another person.

The sequence 2-5-8-4 was judged much less aversive than the sequence 2-5-8. Yet the first sequence contains the second entirely and adds one more segment.

In the same study, an unweighted combination of peak discomfort and discomfort at the end explained 94% of the variance in the ratings. Adding duration raised the explained variance by only a further 3%. The calculation on this page is based on this simple mean too.

Peak-end estimate = (score of the worst moment + score of the last moment) ÷ 2.

The studies on this page measured painful experiences. That is why peak here means the worst moment. In the demo too, I take the peak as the worst moment and read the number as a rough estimate.

Live example: the feeling curve of an order

The flow below belongs to a fictional online order. I gave each of the six moments a feeling score between 0 and 10. Use the three toggles to improve different parts of the flow, and see how the two numbers change.

The feeling curve of an order

• Worst moment
• Last moment

1. 1. Search8 / 10
   Search brings up the right product on the first try.
2. 2. Cart7 / 10
   Shipping costs are already shown in the cart.
3. 3. Address5 / 10
   Street, city, state and ZIP code are typed by hand, and the form is long.
4. 4. Payment2 / 10
   Worst momentThe card is declined. No reason is given, and the card fields are cleared.
5. 5. Waiting5 / 10
   The “Processing your order” screen spins for a few seconds.
6. 6. Confirmed4 / 10
   Last momentIt only says “Order received”. No number, no delivery date.

Mean of all moments

5.2

(8 + 7 + 5 + 2 + 5 + 4) ÷ 6 = 5.2

Peak-end estimate

3.0

(worst 2 + last 4) ÷ 2 = 3.0

Starting point: the worst moment is step 4 (Payment, 2 points), the last moment is the confirmation (4 points).

A fictional online order; the steps and points are made up. 0 means very bad, 10 very good. Turn the toggles on and off; the curve and both numbers change together.

You can also check the numbers by hand. Three example calculations:

• Starting point: scores 8, 7, 5, 2, 5, 4. Mean 31 ÷ 6 = 5.2. Peak-end estimate (2 + 4) ÷ 2 = 3.0.
• If only the address step in the middle rises from 5 to 8: mean 34 ÷ 6 = 5.7. The peak-end estimate stays at 3.0.
• If only the worst moment rises from 2 to 5: mean 5.7 again. The worst moment is now the confirmation with 4 points, and the estimate is (4 + 4) ÷ 2 = 4.0.

In both cases, the flow got 3 points better overall and the mean stayed the same. But the peak-end estimate only rose when the worst moment was fixed. If you improve only the end, the estimate rises to 5.0.

The worst moment and the end in the interface

To find the worst moment of a flow, I look at the screen where the person hesitates, goes back or looks for help. In e-commerce, I look at the payment step first. Declined cards and the jump back from the 3-D Secure page happen there.

Payment and confirmation

The end of payment is the confirmation page. In my view, it is also one of the screens that gets the least design attention.

“Order received” on its own isn’t enough. The order number, the expected delivery date and a way to make changes answer the questions on the person’s mind. Details are in the checkout pattern.

Error recovery

An error moment is a strong candidate for the worst moment of a flow. How bad that moment stays is decided by the design.

A message that gives the reason, and fields that aren’t cleared, turn the same error into a lighter moment. Examples are in the error messages pattern.

Cancellation flow

Cancelling a subscription is the last screen of the relationship with the person. A cancellation that drags on or assigns blame spoils the end of months of good use. For an honest cancellation, see the subscription cancellation pattern.

The limits of the rule

The rule doesn’t hold in every case. The authors of the 1993 paper write that duration isn’t always neglected. If an experience lasts much longer or much shorter than expected, duration can play a role too.

According to the authors, peak and end aren’t the only factors either. The speed of an improving or worsening trend also affects the rating. They also suspect that in some cases the first moments dominate the memory.

So don’t read duration neglect as “the flow can be longer”. Cutting waiting times is still your job. The rule tells you to watch two points alongside duration: the worst and the last moment.

How to use it in your own flow

1. Break the flow into steps. A sketch as a user flow is enough for this.
2. In testing, ask at each step how the person feels. The feeling curve in a customer journey map is made for exactly this.
3. Mark the lowest and the last point of the curve. Start your list of fixes with these two points.
4. Treat the last screen as a design task of its own. Also draw the confirmation and the situation after an error separately.
5. After the change, ask the person to rate the experience as a whole. How to set up the session is in the usability testing guide.

My note

In one sentence, the rule sounds easy: make the end nice, and the rest is forgotten. That isn’t how I read it. The formula also contains the worst moment, and in my view fixing it is worth more than decorating the end.

The 1993 paper also has an ethical discussion: adding pain at the end purely for a better memory raises the question of informed consent. In the interface, the equivalent is covering a bad flow with a shiny thank-you screen. If the person runs into the same error on their second order, that screen no longer helps.

If you want an outside look to find the worst moment of your flows, the UX Audit Pro checks up to ten flows of your product from start to finish.