Retention Curve

A thousand customers join a service in January, six hundred are active in month three and five hundred meet the same rule in month twelve. Plotting each month's share of the original thousand shows how retention changes over time.

Amplitude describes retention curves and cohort analysis, while Mixpanel documents retention reporting choices, and their tools may handle dates and returning activity differently, so a curve should name the exact method instead of treating every chart as interchangeable. Define the cohort start event, whether signup, first purchase or first meaningful use, because these create different baselines.

Choose the activity that means retained, since a paid subscription, completed transaction and app login are not equivalent, and specify the interval of day, week or month, because a weekly-use service may look weak on a daily return measure. For a simple month-twelve point, 500 retained out of 1,000 starting customers gives 50%.

Use the original cohort as the denominator at each point under this basic method, and do not replace it with the previous month's survivors. Some products report unbounded or rolling retention, where returning on or after a period counts, and that variant should be labelled because it differs from exact-period retention.

A customer can be inactive one month and return later, so under an activity-per-period definition the plotted line may rise as well as fall, whereas a survival curve for continuously subscribed customers usually declines or stays flat, which is why the rule should be explained before interpreting its shape. Use mature cohorts, since a group that joined last week cannot have a valid month-twelve point.

Exclude test users and duplicated accounts under consistent rules, because tracking errors can make a curve look healthier or worse, and mark product changes that alter event collection, since a renamed login event can create an artificial drop. Compare cohorts started in different periods to see whether onboarding or product changes relate to retention, but do not assume a better-looking cohort was caused by one feature, as marketing mix, pricing and season can also shift.

Segment by plan, acquisition route or customer need, because an average curve can hide a valuable group with a sharp early drop, and protect privacy when a cohort is small since a handful of known accounts can be identifiable from granular reporting. Look at the early slope, as a steep decline after signup may suggest poor onboarding or mismatched expectations, and at the later portion, where a flatter line can suggest a stable core but not that those customers will stay forever.

A line that reaches a small plateau may still represent little revenue if the survivors are on low-value plans, so pair customer retention with revenue retention where relevant, because existing accounts may grow or shrink even if account count stays constant. Consider cancellation timing and billing grace periods, since a payer classified as active can differ from a user who actually uses the service, and check the return threshold because counting a notification-open as active can flatter retention without real value.

Report the cohort size alongside percentages, as a 50% rate from ten customers is much less stable than from thousands, and use confidence and practical judgment when comparing small differences because a graph alone does not establish a meaningful change. Interview customers or examine support feedback for reasons behind the curve, update the chart consistently after data corrections since historical points may change when late events are added, and remember that the curve is most useful for seeing when customers stop finding enough reason to return and where the business should investigate.