Data
Analyse cohorts and funnels
Where people drop out, by when they joined.
Fill it in
Signup, activation and usage events.
The moment that predicts staying.
By cohort, so improvement is visible.
Your prompt
Analyse cohorts and the funnel. Activation: [your activation] Cohort by SIGNUP period and read down the columns. An overall retention number mixes people who joined last week with people who joined last year, so a product that is genuinely improving looks flat and one that is decaying looks fine. Draw the curves and say whether they FLATTEN. A curve that flattens has a real retained user base; one that keeps declining does not, and no amount of acquisition fixes it. Separate the funnel by cohort too. A drop-off that appeared three months ago is a change you shipped, and the aggregate hides the date.
Use Analyse cohorts and funnelsOpens with everything above already filled in.
Why this works
An overall retention number mixes people who joined last week with people who joined last year, so a product that is genuinely improving looks flat. This cohorts by signup period, and says whether the curve flattens, which is the difference between a user base and a leak.