Start with event definitions, not charts
Activation analysis is only useful if “started onboarding”, “connected data”, “first key action” and “activated” mean the same thing to Product, Marketing and the data layer. Before analysing performance, I validate event triggers, user IDs, timestamps, environment filters and expected ordering.
Why this mattersA beautiful retention chart built on unstable events is still wrong. Event taxonomy and data-quality checks are part of the analysis, not a separate technical task.
Activation funnel
The largest absolute loss occurs before users connect their data. That is the first place I investigate qualitatively and quantitatively rather than trying to optimise every step at once.
Cohort retention
Later cohorts retain better. The next question is not “great, retention is up” but what changed? Release history, acquisition mix, onboarding speed and feature adoption are all candidate drivers.
Technical proof
The cohort logic starts with a stable user grain, assigns each user to a signup cohort, then calculates active weeks relative to that cohort. The event contract sits upstream of the chart.
UserStable user ID + signup
EventsNamed behavioural actions
CohortSignup week / month
RetentionActive users by relative week
DATE_DIFF(
activity_week,
cohort_week,
WEEK
) AS week_number
COUNT(DISTINCT user_id) AS retained_users
Representative synthetic logic. The event names and dataset are created for this portfolio.
Segment signal
| Segment | Activation | Retention |
|---|
Hypothesis worth testingUsers who invite a teammate have materially stronger activation and retention. That could indicate collaboration creates value, or simply that high-intent customers invite teammates. That relationship should be tested causally before becoming a product recommendation.