A cohort can look healthier because the business acquired a different mix of customers, not because the experience improved. Comparing two monthly groups without examining their composition can turn a change in acquisition into an apparent change in retention. The practical task is to preserve both the overall business result and the underlying customer differences.
This guide proposes a cohort review for commerce teams. It does not provide benchmark retention rates or assume that a particular repeat-purchase pattern is appropriate for every product.
Define entry and observation consistently
Choose what puts a customer into a cohort: first paid order, first delivered order or another explicit milestone. Use one definition across the comparison. A first order that is later canceled may need separate treatment from a completed customer relationship.
Then compare equivalent elapsed periods. A customer acquired last week has not had the same opportunity to reorder as one acquired three months ago. Show the maturity of each cohort and avoid filling unobserved future periods with zeros. Forecasts, if used, should be visibly separate from observed behavior.
Describe the acquisition mix
Record characteristics that plausibly affect the expected journey: first product category, offer type, acquisition source, geography and whether the initial purchase was a gift or a replenishable item when that information is legitimately available. Do not collect sensitive attributes simply to create more segments.
| Mix change | Possible interpretation risk |
|---|---|
| More durable-product buyers | Lower short-term repeat may reflect product cadence |
| More discounted first orders | Initial economics and later behavior may differ |
| More gift purchases | Recipient and purchaser journeys can diverge |
| New delivery market | Service experience may change independently of marketing |
| Different first-product category | Replenishment opportunity may not be comparable |
These are hypotheses to investigate, not automatic explanations. Product type does not excuse a poor experience; it helps define what a useful outcome would look like.
Keep total and like-for-like views together
The total cohort describes the customers the business actually acquired. A like-for-like view compares a more stable subset. Both are useful. Showing only the stable subset can hide a damaging acquisition shift; showing only the total can hide a real improvement within segments.
For example, suppose the overall repeat rate falls while each comparable product group remains similar. A larger share of naturally infrequent purchases may explain part of the aggregate change. That observation still raises a commercial question: is the new acquisition mix intentional and economically acceptable?
Where sample sizes permit, compare outcomes within a few meaningful groups and show their relative weight. Avoid creating dozens of tiny cells that appear precise but contain very little evidence. Unknown classifications should remain visible rather than being allocated to whichever group makes the report cleaner.
Add economics and experience
Repeat purchase alone does not establish customer quality. Review contribution, returns, support demand and fulfillment problems alongside repeat activity. A group that reorders frequently only under heavy discounts can differ economically from a smaller group that buys at a sustainable price.
Do not let one high-value order dominate the interpretation without showing concentration. Likewise, averages can hide customers with very different experiences. The purpose is not to create a composite score that conceals tradeoffs, but to make the tradeoffs reviewable.
Write a decision note with alternatives
A useful note states the observed change, how much of it appears related to mix, what remains unexplained and what action is justified. Possible actions include repairing a first-use problem, changing acquisition targeting, revising the offer or waiting for a cohort to mature.
The prospecting review guide connects acquisition quality to budget decisions. The first-month customer uncertainty map can help investigate experience differences after the first order. Cohort analysis becomes useful when it explains the population being compared, instead of treating every month’s customers as interchangeable.