Cohorts

Saved audiences built from profile properties and behavior. A cohort keeps evaluating, so it is always the current group, and it is the audience unit across Drill, Profiles, targeting, and adaptive variables.

What a cohort is

A cohort is a saved audience: every profile that matches a definition you write from profile properties and activity. "Highly engaged" might be profiles with at least ten sessions in the last 30 days. "Slipping away" might be profiles with no session in the last 30 days. "Enterprise on iOS" might be profiles whose plan property is enterprise and whose platform is iOS.

Unlike a one-off filter, a cohort keeps evaluating. Profiles enter it when they start matching and leave when they stop, so a cohort is always the current audience, and the section records how its size moved day by day.

That makes a cohort the unit of audience across Countly. You can filter Drill and Profiles by membership, target a push or a journey at a cohort, switch adaptive variables on for it, and compare cohorts against each other or against everyone else.

Two kinds of audiences share the machinery. Cohorts, built here, are rule-based. Profile groups are hand-picked lists managed in Profiles; they appear as membership options in the builder and in the same pickers.

How membership works

Membership is stored on the profile. When a profile matches the definition it is marked as a member with an entry time; when it stops matching the mark is cleared and an exit time is recorded. The detail page's charts are built from those transitions: one series for how many profiles were members at the end of each day, one for how many entered and exited.

With real-time cohorts on (the default), each incoming session or event is checked against every cohort's activity criteria, so a profile joins "performed Purchase in the last 7 days" moments after the purchase. Property conditions are re-evaluated as profile properties change. Rolling windows such as "in the last 30 days" expire on the server's regeneration schedule, so a profile whose last purchase ages out leaves the cohort at the next run rather than at the exact minute.

The state pill next to the name says where the cohort is. Live means membership is being maintained. Calculating means the first pass over existing profiles is still running, and the count grows until it finishes. Periodic appears when the server evaluates cohorts on a schedule instead of in real time.

Editing a definition starts over. The old membership marks and the history are dropped and the cohort is rebuilt from scratch, because the new rule may include or exclude anyone. Recalculate does the same on demand without changing the definition, which is the fix when a count looks stale.

Use Cases

Define a behavioral segment

Pick out the profiles whose behavior you care about, such as "at least 10 sessions in the last 30 days", so the rest of Countly can work on that group.

1. Open Cohorts and click Create cohort.
2. Name it. A name that states the rule ("10+ sessions, 30 days") stays readable in pickers across the product.
3. Under Activity criteria, add a step: performed Session at least 10 times in the last 30 days. Add a filter on the step to narrow it to one platform or version when the segment is platform-specific.
4. Under Profile criteria, add property conditions when the segment also depends on who the profile is, such as country or a custom plan property.
5. Choose the visibility and save. The cohort shows Calculating while existing profiles are evaluated, then Live.

Catch profiles slipping away

See churn before it is final. Profiles who used to be active but have not been seen recently are the audience for a win-back message and a number worth watching.

1. Create a cohort with an activity step did not perform Session in the last 30 days, and a profile condition First seen older than 30 days so brand-new profiles do not count as lapsed.
2. Open the cohort. Number of profiles in cohort shows the lapsed audience growing or shrinking, and entering on the flow chart is the daily churn signal.
3. Run Event divergence to see what these profiles did less of before they went quiet. Events with a strong negative divergence are the moments to protect.
4. Target the cohort from Push or Journeys. Profiles who come back leave the cohort on their own, so exiting on the flow chart is your win-back rate.

Learn what makes a cohort different

Explain a cohort, not just count it. The detail page compares members against everyone else on three axes.

1. Event divergence ranks every custom event by how much more (or less) of the cohort performs it than the rest, with a frequency column for how much more often performers do it. Click Run analysis; it is the heaviest query on the page and runs on demand for the selected date range.
2. Breakdown of profiles compares the top values of up to three profile properties (platform, country, app version, custom properties) between members and everyone else. Use Configure to pick the properties.
3. Retention divergence draws classic retention for members against the rest at daily, weekly, or monthly stops over a fixed window, so you can see whether the cohort sticks around better.
4. Any divergence row hands off to Drill with the membership filter and the event already filled in, for the profile-level view.

Read the three together. A cohort that performs "Export" three times as often, skews to desktop, and retains better is a segment worth designing for.

Compare two audiences over time

Put cohorts side by side, such as this quarter's signups against last quarter's, or iOS against Android.

1. On the list, click Compare, select two to four cohorts, and confirm.
2. Choose the metric for the main chart from its title: total profiles, total sessions, or average session duration.
3. Set the date range. Turn on comparison in the date picker to overlay the previous period.
4. Read Profiles entering and Profiles exiting to see which audience is growing, and the detailed comparison table for the exact numbers.
5. Save it from the page icon beside the title with Save shortcut, to reopen the same comparison, metric included, from the list's saved comparisons.

Put a cohort to work

Use the audience, not just look at it. A cohort is available wherever Countly asks "who".

1. Drill and Explore: filter any query by membership, or break a metric down by cohort to compare members against non-members. A step in the builder can also require membership in another cohort.
2. Profiles: filter the profile list by cohort, or open the member list from the cohort's action menu and export it from there.
3. Push, Journeys, Surveys, and in-app content: pick the cohort as the audience. Membership is read when the message is sent or the journey step runs, so the audience follows the rule.
4. Adaptive variables: switch a variable on for one cohort to ship a tailored experience.
5. Retention and Funnels: scope the analysis to a cohort.
6. Hooks: trigger a webhook or an email when a profile enters or exits a cohort.
7. Dashboards: cohort-scoped tiles keep the audience definition in one place.

Every consumer reads the same membership, so the audience a journey targets is the audience Drill counts.

FAQs

Why is my cohort stuck on Calculating?

The first pass evaluates every existing profile, and large projects take minutes. If the count stops growing but the state never changes, run Recalculate from the action menu. A cohort that always reads Periodic is on a server that evaluates cohorts on a schedule rather than in real time.

Why did the history reset after I edited the definition?

Because the new rule may include or exclude anyone, an edit clears the membership marks and the daily series and rebuilds the cohort from scratch. Duplicate the cohort first when you want to keep the old one's history.

Why do the cohort's size and a Drill count differ?

A cohort counts profiles that match the rule now. Drill counts profiles with matching events in a date range. A profile with a purchase last month is in "purchased in the last 30 days" until the window passes, but does not appear in a Drill query over this week.

What is the difference between a cohort and a profile group?

A cohort is a rule; profiles enter and leave as they match. A profile group is a hand-picked list managed in Profiles. Both are audiences in the same pickers.

What happens when I delete a cohort that a push or a journey targets?

The membership marks are removed and the audience is gone. Retarget messages, journeys, and adaptive variables that use the cohort before deleting it.

How do I export the members?

Open the member list from the action menu (or click the total in the size chart) and export from Profiles. The list page's Download exports the list of cohorts, not their members.