The Peak-End Rule Meets Onboarding
People remember the best moment and the last one, but the evidence has narrow edges.
Pennant, a scheduling tool for clinics, runs a nine-step setup. Steps one through seven are smooth. Step eight is a calendar import that fails for about a fifth of users and dumps them into a support article. Step nine is a success screen with confetti. The team's NPS from new accounts is poor and nobody can see why, because seven of nine steps are fine and the completion rate is 91 percent.
The averaged experience is good. The remembered experience is a failure followed by confetti, which reads as mockery.
What gets remembered from a setup flow
What the research actually shows
Kahneman and colleagues ran a cold pressor study published in 1993. Participants held a hand in painfully cold water for 60 seconds, and in another trial for 60 seconds plus 30 more seconds during which the water was warmed slightly but stayed uncomfortable. The second trial contained strictly more discomfort. A majority chose to repeat it. Redelmeier and Kahneman found a similar pattern in 1996 with patients recalling colonoscopies: retrospective ratings tracked the worst moment and the final moment far better than the total duration.
That is the finding. Notice its shape. It concerns remembered evaluation of a single bounded episode of physical sensation, measured shortly afterwards. Onboarding is a series of episodes, spread over days, mixed with expectations the user brought from a competitor and from your marketing site. Applying peak-end to it is an extrapolation. It is a reasonable extrapolation, and the lab result on duration neglect is solid, but if you tell an executive that research proves onboarding should end on a high, you are overselling.
Three handles the research gives you
- Duration neglect
- Total length matters less to memory than intensity at key moments. A long setup is not automatically a bad memory. A short one with a sharp failure often is.
- The peak
- The most intense moment dominates recall, and negative peaks tend to be sharper than positive ones. Your first job is removing the worst moment, not adding a better one.
- The end
- The final moment gets weight out of proportion to its length. Whatever the user does last is disproportionately what they will tell a colleague about.
For Pennant the order of work is now obvious. Fix step eight first. Confetti on top of a broken import is a bigger insult than no confetti. Only once the peak is flattened does the end become worth investing in.
An end that celebrates the product
- You're all set up! Welcome to Pennant.
- Setup complete. 9 of 9 steps done.
- Your account is ready. Explore the dashboard.
An end that hands over something real
- Your first week is live. 14 appointments imported, 3 need a room assigned.
- Assign rooms (3) and your Monday is ready to print.
- Here's Monday. Everything below is editable.
The right column ends inside the user's work rather than inside your funnel. The last thing they see is their own clinic's Monday, not a badge for finishing your form. Kathy Sierra's argument in Badass is the useful version of this: the goal is for the user to be good at their job, so the finish line should be a piece of their job that is now done.
Worked example
Rescuing the failed import
Pennant cannot make every calendar import work. So they change what failure looks like. Instead of an error and a link to docs, the step falls back to: Import didn't work for this calendar. We've added your 3 recurring clinics manually so you can keep going, and we'll email you when the connector supports Outlook 2016. The peak is still negative, but it is now a shallow dip inside a path that continues, rather than a wall.
One measurement note. Peak-end is about remembered experience, so instrument it that way. A survey during setup measures the moment. A survey three days later measures the memory, and the memory is what drives renewal conversations and referrals. If those two numbers diverge, the gap is telling you where your peak is.
Quick check
Why should Pennant fix the failing import before improving the success screen?
The takeaway
Remove the worst moment before you decorate the last one, and end onboarding inside the user's real work.
Try this tomorrow
Instrument the single highest drop-off or error step in your onboarding this week and rewrite its failure state so the user can continue rather than stop.
Answer the check above, then bank the day.
Where this comes from
- When More Pain Is Preferred to Less: Adding a Better End (1993), Kahneman, Fredrickson, Schreiber and Redelmeier
- Patients' memories of painful medical treatments (1996), Donald Redelmeier and Daniel Kahneman
- Thinking, Fast and Slow, Daniel Kahneman
- Badass: Making Users Awesome, Kathy Sierra
User Psychology is one of six tracks. These lessons summarise and build on the work above, they do not reproduce it. Buy the books, they are better.