Journal · 8 min · N. Okoye
When cohorts flatten the story
Cohorts feel rigorous because they have a start date. That date is often an install, a first open, or a first session — three different doors into the same building. Average them and you get a calm line that describes nobody.
Acquisition week is a filing cabinet
People who arrived in the same week do not share a motive. One saw a performance ad. One was forwarded a login. One reinstalled after a phone swap. If your App Analytics default is “users who did X in week W,” write down which X you used. Then split once by a motive you can observe: paid versus organic, or invited versus self-serve. If you cannot observe motive, do not pretend the week is a segment.
Delayed onboarding
Some products collect an account on day one and deliver value on day twelve, after a human reviews a document. A D7 cohort will call those people lost. They are waiting. In Retention Signal Mapping we ask for the operational SLA next to the curve. If the SLA is ten days, D7 is a measure of impatience in the analytics team, not of the product.
Platform mix
iOS, web, and a desktop leftover in the same cohort will produce an average that no engineer can act on. Start by splitting platform, even if the sample gets ugly. Ugly and true beats smooth and unusable. The mixed-platform adoption essay on this journal continues that argument with feature flags.
None of this requires a new vendor. It requires a refusal to present the first cohort the tool offers.