03 The wrong question
Most onboarding checklists assume you know what activation is.
Most operators don’t. They have a guess. The guess is usually “completed account setup” or “invited a teammate” or “reached the demo dashboard.” The guess feels right because it’s the first thing the product asks the user to do. That’s exactly why it’s usually wrong.
The activation event is the leading indicator with the highest predictive power for 90-day retention. Not the first action. Not the most-tracked action. The action that, when a user takes it, predicts they’ll still be using the product three months later. That’s a regression problem, not a vibes problem.
How to find yours
Pull a cohort of 90-day-retained users and a cohort of 90-day-churned users. For each cohort, list the early-life behaviors (first 14 days). Find the behaviors where the two cohorts diverge most. The behavior with the largest delta — usually the first time a user does the thing the product is actually for — is your activation candidate. Validate with a holdout cohort and check that the relationship isn’t spurious (does it survive controls for plan tier, acquisition channel, and usage frequency).
At the DTC SaaS in the case study, the guessed activation was “completed signup.” The real activation was connected store + imported first product. Two distinct events in sequence, and the gap between them was where 60% of trial users died. The welcome series was firing happily into the gap. The email program looked great. The activation rate was terrible. That’s the signature pattern: a working email system optimizing the wrong milestone.
Validating the metric
The first cut from the regression gives you candidates, and the validation step most teams skip is building a matrix for each one. Rows are the number of times the action is performed. Columns are time periods (24h, 48h, 7d, 14d, 28d). Cells contain correlation with 90-day retention plus sample size, positive predictive value, and negative predictive value.
Three filters narrow the candidate set. Drop any cell with correlation below 0.3 because thats a weak signal pretending to be one. Drop any cell with sample size too small to be statistically meaningful. Then look for the inflection point, where adjacent cells have correlations that jump significantly between them, because that inflection is usually where your real activation metric lives. Among the surviving candidates, pick the one thats simplest to state in one sentence and most operationally clear, because the metric anyone in the company can recite from memory is the one the team will actually optimize against.
After identifying a correlated metric, run an experiment that moves users to hit it. If retention rises proportionally with users hitting the metric, the relationship is causal and youve got your activation event. If retention doesnt move when the metric does, your correlation was incidental and you go back to the regression. Correlation never proves causation by itself, the validation experiment is what resolves which one you have, and skipping it leaves your lifecycle program optimizing toward a number that doesnt actually drive retention.