Cohort tables invite horizontal comparison. Colour makes the invitation stronger. Yet adjacent rows may contain users with different acquisition intent, product eligibility or opportunities to return.
Four checks before interpretation
First, confirm that every cohort has completed the observation window. A recent row cannot fairly compete at week eight. Second, inspect acquisition mix; a campaign or partnership can change who arrived without the product changing.
Third, review identity rules. Login changes, reinstall behaviour and account merges can alter whether a returning person is counted as retained. Fourth, check eligibility: if a feature or recurring task is available only to part of the population, whole-base retention can conceal the relevant effect.
Prefer explicit exclusions
Write exclusions next to the chart: internal accounts, test traffic, unsupported versions, migrated users or regions with a different launch date. This makes the denominator discussable and prevents each stakeholder from imagining a different population.
Comparability is not a preliminary nuisance. It is part of the result. Sometimes the most responsible conclusion is that a historical tracking change prevents a clean before-and-after claim.