Two questions need two denominators
Activity retention commonly creates a cohort at first use or first value action, then observes whether people are active again on Day 1, Day 7, and Day 30. Repeat purchase begins with a first order and looks for another completed order within a later window.
Using all visitors as the repeat-purchase denominator, or first purchasers as the ordinary activity-retention denominator, makes the measure answer a different question from its label.
A hypothetical cohort
This is a hypothetical work-note scenario, not real business data. A cohort contains 10,000 people who completed a first value action. Of them, 1,800 are active again on Day 7, giving Day 7 retention of 18%. A repeat-purchase analysis would use a separate first-order population and order definition.
| Observation | People | Rate |
|---|---|---|
| First value action | 10,000 | 100% |
| Active again on Day 7 | 1,800 | 18% |
Fix four definitions
- Cohort entry eventRegistration, first use, first value action, and first order are not interchangeable.
- Return or purchase eventOpening the product, completing a key action, and paying for an order represent different behavior.
- Time windowCalendar days, rolling hours, and complete weeks assign boundary users differently.
- Identity and deduplicationCross-device sign-in, refunds, cancelled orders, and duplicate payments need consistent treatment.
Do not compare immature observation periods
A user entering the cohort today has had no chance to contribute Day 30 retention. Cohort comparisons should include only people who have completed the relevant window. The same applies to repeat purchase: no purchase yet is not a final outcome before the window closes.
Separate frequency from value
Repeat-purchase rate shows how many people buy again, not their order count or spend. Order frequency, average order value, and refunds may add context, but each needs its own definition instead of being merged into a vague measure of user value.