Skip to content
AdPixDocs
Analytics
GuidesDevelopers
Search the docsEnglishOpen console

Cohort retention

This report is not ready yet — it is parked in the Coming soon group. It does build a retention grid, but the cohorts are derived from the selected date range and the day boundaries are not evaluated in the property's timezone.

GuideSoon3 minUpdated

Why it is still Coming soon#

The item is called Cohort retention in the side navigation, sits under Coming soon, and carries a Soon badge. It opens and builds the grid, but three behaviours in the calculation make conclusions drawn from it unsafe:

  • Cohorts are derived from the selected range. A user's acquisition date is the first day an event of theirs appears inside that range. Someone whose real first visit predates the range is assigned to a cohort inside it. The result is an optimistic retention rate on short ranges.
  • Days and weeks close in UTC. Every finished report buckets time in the property's timezone. This grid does not.
  • Retention is not definable. Returning means any event at all from that user in that period. You cannot tie it to a purchase or a key event, apply a segment, or export the grid.

What the page shows if you open it#

A selector with Weekly and Daily, and a card named after that choice — Retention by acquisition week or Retention by acquisition day. Each row is one cohort: its acquisition date, a Users column holding the cohort's starting size, then the period columns, labelled with W for weeks or D for days, from period zero through period twelve. Each cell is the percentage of that cohort seen again in that period, shaded darker as the percentage rises. Hover a cell for the raw user count. With nothing to show you get No cohort data in range.

The grid always shows twelve periods, and period zero is the cohort's own starting size, so it is always one hundred per cent.

What covers this today#

There is no complete replacement yet, but these three answer part of the same question with numbers you can trust:

  • Explorations to build the table yourself. Put the date and first-touch channel dimensions next to the user, key-event and key-event-value metrics, and compare consecutive ranges. It is not cohort retention, but it gives you a return trend on your own definition.
  • Conversion funnel when the real question is how many took the next step, rather than how many came back.
  • Engagement for returning versus new users and real time on site.
Choose a long range

If you do work with this grid, select a range far longer than the cohort lifetime you care about. The longer the range, the fewer users are misclassified as new. How the range selector behaves is covered in Date ranges and comparison.

Frequently asked questions#

Why does retention look unrealistically good on a short range?

Because each user's acquisition date is the first day they were seen inside the selected range. A customer who first arrived months ago counts as a new member of this week's cohort if they come back this week, which inflates the later columns. Pick a longer range to reduce the error.

What does retention mean here?

That the user produced any event in that period — a single page view is enough. You cannot base retention on a purchase or on a specific key event, and that is one of the reasons the report is not finished.

Are day boundaries evaluated in my local time?

No. Every other report evaluates its range and its day, week and month buckets in the property's timezone; this grid closes days and weeks in UTC. For an Iran-based property that means a few hours of drift in which cohort a user lands in.

Build with the APIUnderstand where revenue comes from.
Was this page helpful?