Customers do not remember the average of their experience with your product. They remember the worst moment and the last moment.
This is the peak-end rule, and it changes where you should invest for retention.
What the peak-end rule actually says
Behavioral economist Daniel Kahneman ran experiments showing that people judge experiences based on two data points:
- The peak: the most intense moment (positive or negative) during the experience
- The end: how the experience felt when it concluded
Not the average. Not the duration. Not the sum of moments. Just the peak and the end.
His most famous experiment involved colonoscopies: patients rated the experience worse when the doctor stopped mid-procedure (bad ending) than when the procedure was extended by 60 seconds of mild discomfort (worse average but better ending). The extended-but-gentle-ending group would agree to another procedure. The abruptly-ended group would not.
Apply this to SaaS retention.
What this means for your product
Fix rare bad moments, not average moments
Most product roadmaps invest in improving average experience: faster loading, cleaner UI, better copy. These matter, but they compound slowly.
A single bad moment that affects retention:
- An outage that hit at the wrong time
- A data loss that took a customer's work
- A billing error that created legal or compliance risk
- A feature deprecation that broke a workflow
- A support experience that was dismissive
These bad moments are what customers tell others about. And what they remember at renewal.
Roadmap investment in preventing rare-but-devastating moments has better retention ROI than investment in making average moments incrementally better.
Design the "peak experience"
Also design a positive peak. The moment your product does something genuinely delightful should be memorable. Slack's "here's a bunny that means everything is fine when the search is broken" is an example. The customer forgives search being broken because the recovery moment was warm.
What this means for support
Peak-end applies to support interactions strongly.
The last email matters more than any previous email
A 10-email support thread that ended with a curt "issue resolved, closing ticket" leaves the customer feeling worse than a 10-email thread that ended with a genuinely helpful, personalized close.
Train support to end well. The final message should:
- Confirm what was resolved
- Acknowledge the frustration if it was a long thread
- Offer a proactive next step ("If X happens again, here's how to skip the queue")
- Sound like a human wrote it
The recovery from a bad support moment
When a support interaction goes wrong, the recovery becomes the peak. A CSM who calls to personally apologize and fix a botched ticket creates a positive peak that outweighs the original negative moment.
This is why customers who complain and get great recovery often become better advocates than customers who never had an issue.
What this means for cancellation flows
Cancellation is the last moment of the relationship. It becomes the peak of what customers remember.
Graceful cancel flows increase win-back rates
Customers who cancel gracefully (respectful save flow, clean confirmation, useful goodbye email) come back 8-15%. Customers who cancel through a friction-filled flow come back 2-5% and actively tell others not to buy your product.
The last email in the relationship matters most
Your cancellation confirmation email is often the last thing they read from you. It should feel like a relationship ending well, not a transaction closing. See how to write a cancellation email.
What this means for onboarding
The peak-end rule applies to individual sessions too. Users' memory of any session is dominated by:
- The peak moment during the session (found something valuable, or hit friction)
- How the session ended (task completed, or gave up)
Sessions that end mid-task feel worse than sessions that end at a natural completion, even if the total time and effort were the same.
Design onboarding so users can end at natural moments. Progress indicators help, but so do "you're at a stopping point" cues.
The counterintuitive implication
You may be better off having some rare bad moments in exchange for consistent peak experiences. Trying to be flawless everywhere often produces flat, unmemorable products.
Snowflake, Notion, and Linear all have specific frustrating limitations that users complain about. They also have specific peak moments that users evangelize. Neither eliminates the other, but the peak moments create the retention.
How to apply this framework
- Audit your bad moments. Where do customers report the sharpest negative reactions? Fix those first, even if they are rare.
- Design at least one peak moment. Every user should experience something genuinely positive that becomes their memory of your product.
- Redesign endings. Every interaction ends somehow. Support tickets, onboarding sessions, cancellation flows, quarterly business reviews. The ending is what gets remembered.
- Measure by memorable moments, not averages. An 80% average CSAT with one 20% CSAT interaction is often remembered as "the 20% moment" if it was recent enough.
Related concepts
- Customer Effort Score (CES) - how effort creates negative peaks
- How to write a cancellation email - the last-moment application
- What is an aha moment? - designing peaks in onboarding
To score your retention setup for peak-end thinking, take the 60-second Churn Health Check.