Back to all posts
Analytics7 min read·October 25, 2026

How to Read Mobile App Retention Cohorts

TB
ThynkBlox Apps Team
App Engineering

To read mobile app retention cohorts, group users by when they started, then count how many complete a meaningful action in a later period. Compare groups at the same age, using the same return action and observation window. Separate users with different needs or acquisition sources before treating a change in the overall result as product improvement.

What should count as coming back?

Start with the reason someone uses your app. A return should show that they received value again.

For a meal-planning app, that might mean saving another meal plan. For a booking app, it might mean completing another booking. Opening a notification or reaching the home screen can help explain behaviour, but neither necessarily completes the job the user came to do.

Write a definition before reading the chart:

  • Starting action: the event that puts someone into a cohort.
  • Return action: the useful activity you want them to repeat.
  • Return window: the period in which that activity must happen.
  • Counting rule: each person counts once within a window, regardless of how often they act.

Keep app opens as a separate measure if they help answer another question. Someone who opens the app but cannot finish a task needs different attention from someone who never returns.

If the useful action is still unclear, revisit how to scope an app MVP. A clear core task gives you something concrete to measure after launch.

Which starting date belongs in a cohort?

A cohort is a group of users who share a defined starting condition. For this analysis, choose a starting event and group users by the period in which it happened.

Installation, account creation and first completed task answer different questions. Installation measures from initial arrival. First completed task measures repeat use among people who have already reached that milestone.

Neither definition should silently replace the other. If you start at first success, people who installed but never succeeded are outside that cohort. Keep a separate view of that earlier journey.

Label each row with its starting period and starting event. Use a consistent timezone so events near a boundary follow the same rule.

Then distinguish two clocks:

  • Calendar time: when users started, which helps compare releases, campaigns or seasonal conditions.
  • User age: how long it has been since each user started, which makes follow-up periods comparable.

A calendar week can define the cohort row while elapsed weeks define its columns. State how partial weeks are handled; otherwise, users entering near a boundary may receive unequal observation time.

How do you read the rows and columns?

Read across a row to follow one starting group through later periods. Read down a column to compare different starting groups at the same stage.

Before interpreting any cell, check what it means. “Returned during the second week” differs from “returned at any point after the first week”. The first describes activity within a window; the second describes activity across a broader period.

For a window-based view, use this calculation:

Users who completed the return action in the window ÷ users in the starting cohort.

Keep the starting group fixed. Dividing by only those who returned in the previous window answers a different question.

Also check whether the window has finished for everyone being compared. A newer cohort with incomplete follow-up is not ready for the same judgement as an older cohort. Mark unfinished cells as pending rather than treating them as zero.

How often should users need your app?

Choose a return window that matches the intended task.

For an app designed around a daily routine, daily windows may be useful. For a weekly planning task, a weekly view may be more appropriate. For an occasional booking or renewal, start by asking when another need could reasonably arise.

Do not choose the window simply because it is the dashboard default. Write the expected pattern as a product assumption, then examine whether observed use supports it.

Consider measuring both return activity and task completion. A person might return repeatedly because a task remains unfinished. Counting those visits as repeated success would hide the problem.

For an occasional-use product, retention alone may not answer whether the app worked well. You may also need to examine whether people completed their task, whether they needed help, and whether they chose the app when another relevant need arose.

Can an average hide what is happening?

Yes. An overall result combines the behaviour of each group with the number of users in that group.

Illustrative example only: imagine two fully observed starting cohorts in a weekly planning app. Everyone enters after saving their first plan. A return means saving another plan during the second elapsed week. The counts below are invented to explain the calculation, not to provide a benchmark.

Starting cohortAcquisition sourceStarting usersUsers who returned
Earlier cohortReferral8048
Earlier cohortPaid campaign204
Later cohortReferral2012
Later cohortPaid campaign8016

The earlier cohort has 52 returning users out of 100. The later cohort has 28 out of 100.

That looks like a decline. Yet the result within each source is unchanged: 48 out of 80 is the same fraction as 12 out of 20, and 4 out of 20 is the same fraction as 16 out of 80.

What changed was the acquisition mix. The later cohort contained more users from the source with fewer returns relative to its starting group.

This does not establish why the sources differ. It tells you where to investigate before blaming a product change.

Which users should you compare?

Start with a broad cohort view, then separate groups where there is a clear reason to expect different conditions.

Useful comparisons may include acquisition source, intended use, onboarding route or app version at entry. Choose the distinction because it answers a question, not because the dashboard offers it.

For example, a campaign promising quick setup may attract people with different expectations from a referral explaining the full workflow. Review what each group was promised alongside what they experienced.

Avoid creating so many slices that every conclusion rests on a handful of people. Show the starting count beside each result and treat small-group differences as prompts for investigation.

Record what changed between cohorts. If onboarding and acquisition both changed, the chart alone cannot tell you which change explains the outcome.

What common mistakes should you avoid?

Use this checklist before sharing a retention conclusion:

  • Counting any open as success: confirm that the return action reflects useful activity.
  • Comparing unequal follow-up: wait until equivalent windows are complete.
  • Changing the denominator: keep the starting cohort fixed for this view.
  • Mixing definitions: label window-based and cumulative return measures clearly.
  • Reading only the average: inspect relevant acquisition groups and their sizes.
  • Treating a release comparison as proof: investigate other differences between cohorts.
  • Ignoring measurement changes: check whether an event was renamed, duplicated or stopped recording.

Test the measurement journey alongside the user journey. Our guide to testing an Android app before release provides a useful starting point for checking the actions people must complete.

What should you do after reading the chart?

Turn the observation into a narrow question.

If people open the app but do not complete the return action, inspect that task. If one acquisition group behaves differently, review its promise and starting experience. If results change across every group, check shared product changes and measurement before choosing a fix.

Write down the observation, a possible explanation and the evidence that would challenge it. After making a change, compare fully observed cohorts under the same definition.

Include this review in app maintenance after launch, so retention informs a continuing decision rather than becoming a chart nobody acts on.

Frequently asked questions

Should the cohort start at installation or first use?

Use installation to examine the journey from arrival. Use a meaningful first action to examine repeat value after initial success. Keep both clearly labelled if you need both questions answered.

Is a user lost if they miss a return window?

The chart only establishes that they did not complete the selected action during that window. They may return later. Avoid labelling them permanently lost on that evidence alone.

What retention result should we aim for?

Set a goal around your app’s purpose, expected usage pattern and current baseline. The illustrative counts above are not targets. Judge progress using comparable groups and evidence that users are completing the task they came for.

Ready when you are

Planning an app of your own?

Tell us where your project is today. We will help you work out the scope before you commit to a budget.