AddThisFeature

Cohort Analysis

Group users by when they started and watch how each group behaves over time.

involved Analytics

What it adds

A retention grid that buckets users by a qualifying moment and tracks what each bucket does in the periods that follow.

What your agent is told to do

5
  1. 1

    Define the qualifying event that puts a user in a cohort, and freeze their membership at that moment. A cohort that re-sorts itself as users change is not a cohort.

  2. 2

    Support both calendar buckets (users who signed up in March) and rolling buckets (day 0, day 7, day 30 relative to each user's own start).

  3. 3

    Render the result as a triangular grid: cohorts down, periods across, with each cell showing both the count and the rate.

  4. 4

    Read from the app's existing event stream. Event capture is owned by Event Tracking — extend that rather than instrumenting a second pipeline.

  5. 5

    Do NOT restate ordered step-conversion logic here; that belongs to Funnel Analytics. This feature is about time, not sequence.

Edge cases it handles

6
  • A user who qualifies twice must be handled by a stated rule — first qualification only, or a separate row per qualification. Silently double-counting inflates every cell.
  • The most recent cohort has not lived through a full period. Mark partial periods rather than plotting them as a collapse in retention.
  • Late-arriving events change historical cells. Decide whether cells are recomputed or frozen, and say which on the page.
  • Calendar buckets need a stated time zone. UTC and the viewer's local zone put the same user in different months.
  • A cohort with a handful of users produces meaningless percentages. Suppress or flag rates below a minimum cohort size.
  • Deleted or merged accounts must not silently shrink historical cohorts — retention would appear to improve.

Definition of done

8
  • Cohort membership is fixed at the qualifying moment and does not change retroactively.
  • Both calendar and rolling bucketing are available and clearly labelled.
  • Users who qualify more than once are handled by a documented rule.
  • Partial and incomplete periods are visibly marked, not plotted as data.
  • Small cohorts are flagged or suppressed rather than shown as noisy percentages.
  • The bucketing time zone is stated on the report.
  • The feature matches the existing design system.
  • No existing functionality is broken.

Related features

How it works

  1. 1

    Copy the link

    Grab the Markdown instruction URL for this feature.

  2. 2

    Give it to your AI

    Paste it into Claude Code, Cursor, v0, Lovable — whatever you build with.

  3. 3

    It inspects, then implements

    Your agent reads your existing app first, then adds the feature to fit it.

Works with your stack

These instructions are written to adapt. They tell the agent to detect your framework, match your existing design system, and reuse what you already have — rather than assuming a particular stack.

Need it tighter than that? Customize the feature and tell it exactly what you're running.