Foundation · 2 min
What you need to know
How should ecommerce cohorts be created?
Create cohorts from a shared starting event such as first purchase month, first product or acquisition source. Keep the membership rule fixed and compare outcomes at equal elapsed ages.
Should customers be grouped by first purchase month?
Yes. First-purchase month is a useful acquisition cohort because every member begins the customer relationship in the same period. Add source or first-product breakdowns when volume permits.
How do I read a retention cohort table?
Each row represents a starting cohort and each column an equal period since its start. Read across a row to see retention change and down a column to compare cohorts at the same age.
Why are recent cohorts retaining worse?
Recent cohorts may have different sources, promotions, first products, customer fit or first-order experiences. First verify equal-age comparison and consistent identity and retention definitions.
How do I compare retention by acquisition source?
Assign customers to the consistently defined source of their first qualifying purchase and compare equal-age retention, contribution and returns. Show unattributed customers rather than redistributing them.
How do I compare retention by first product purchased?
Group customers by stable first-product or product-family ID, then compare repeat behavior at equal ages. This shows which entry products lead to replenishment, complementary purchases or no further need.
How do I compare retention before and after a business change?
Use cohorts exposed before and after the change at equivalent seasonal periods and ages, while checking acquisition and product mix. A concurrent change or trend can invalidate a simple before-and-after claim.
A quick store check
Build monthly first-purchase cohorts and compare them at equal ages by source and first product.
Foundation · 2 min
Frequently asked questions
Should a customer change acquisition cohort after returning?
No. Keep the original acquisition cohort fixed and use separate behavioral segments for later changes.
Can small cohorts be trusted?
Treat them cautiously; show the count and uncertainty instead of presenting unstable percentages as conclusions.