A single day-seven retention number combines people who arrived for different reasons, through different channels and at different stages of market maturity. When the mix changes, the average changes—even when the underlying experience does not.
Start with the decision
Before splitting cohorts, ask what action the comparison could change. Acquisition source helps when spend allocation is in question. First-use behaviour helps when onboarding is in question. Geography helps when language, payments or local expectations differ.
Keep denominator rules visible
Teams often compare charts built with different definitions of “eligible”, “active” or time zone. Record the cohort entry event, return event, observation window and exclusions beside every curve. In Malaysia, also check whether campaign and festive-period effects alter the users entering a cohort.
Look for durable separation
A one-day gap may be noise. A separation that begins after a meaningful journey step and continues through later windows deserves investigation. Pair the curve with event sequences and qualitative evidence before calling the cause.
The goal is not more segmentation. It is finding the smallest useful comparison that changes a real product or growth decision.