Choose an Amplitude Retention Interval
Match retention intervals to the product’s expected return cycle, then keep the definition stable across comparisons. A weekly workflow can be healthy without daily use, while a daily collaboration task may need a much shorter lens.
Before you start: Have a meaningful start and return event and cohorts with enough observation time. Agree how the team defines a retained user.
Worked example: A monthly planning product should not classify every quiet weekday as a failure. Compare return to the planning action over the interval in which users realistically need it.
Step-by-step guide
Describe expected cadence
Ask when the user needs the product again and why. Use research and task context rather than an arbitrary industry benchmark.
Inspect observed intervals
Review the available usage-interval analysis for the critical event. Check whether different segments have genuinely different task frequency.
Choose the retention definition
Set starting and return events that answer the question. Keep the event meaning distinct from the time interval.
Align cohort maturity
Compare cohorts that have had equal time to return. Recent users should not be treated as failures before the relevant period has elapsed.
Check alternate intervals
Use a second interval as a sensitivity check. If interpretation reverses, explain what each view measures rather than choosing the more flattering chart.
Publish the rationale
Save the cadence, event definitions and exclusions. Record when the metric should be reviewed as the product’s use case changes.
Common mistakes
Daily metric for monthly work
The cadence can misclassify normal behavior.
Incomplete cohorts compared
Newer users have less opportunity to return.
Interval changed silently
The trend can change because measurement changed. Document revisions.
Tips
Write the cadence rationale beside the chart.
Compare equivalent age cohorts.
Separate task-specific retention from generic app activity.
Sources and verification
Based on product documentation checked on 2026-10-24. This is a documentation-based guide, not a hands-on benchmark. Features and availability can change.