Skip to main content
Cohorts

Understand how to create groups of users via Cohorts

Updated over a week ago

What is a Cohort?

Cohort means a group of users all fitting a certain set of criteria within the same time period. One often used cohort is - users who registered for a service on the same date. There can be other kind of cohorts depending on the context, for example:

- Paid vs Organic users (based on acquisition source)

- New (registered in the last 7 days) vs Old (registered more than 7 days ago) users

- Frequent Streamer (streams more than 5 times a week)

Cohort analysis allows companies to identify the behavioral patterns of their customers clearly and adapt and tailor their products and services for specific cohorts.

In Kubit you can:

1. Create and Sample Cohorts

2. Compare how different Cohorts behave, convert and retain

3. Use Cohorts as Filters

4. Export Cohorts to Customer Engagement Platforms such as Braze (add-on feature)

Elements of a Cohort

When building a Cohort anywhere within Kubit important inputs are needed to ensure you see the correct information.

Subject

The unique identifier that will be returned in the Cohort list. This should be relevant for the event/report you are referencing to build the cohort.

  • Ex. A cohort of User IDs that logged in vs. a list of Visitor IDs that visited a page. Likely User IDs aren't available until a user has logged in.

Type

Depending on the type of behavioral filters you want to segment into a Cohort will determine which type you should use.

  • Performed: matches Subjects that performed certain event(s)

    • ex. User IDs that performed the Logged In event

  • Measure: applicable for aggregations or counting of items

    • ex. Count Events of Logged In > 5

      • Users who logged in greater than 5 times

  • Query: Any Subjects whose Query result in the specified time window matches the conditions

    • ex. Give me all User IDs who's result within a Query had a value of > 5

  • Funnel:

    • Converted on step N: Subjects who converted on all steps including step N

    • Dropped off since step N: Subjects who converted on all steps leading up to N and didn't convert on step N

    • Dropped off before step N: Subjects who appear on step 1, then fail to convert on any step before N

      • Best when you want to understand drop off from step 1 of the funnel

  • Retention: any Subject who returned or churned in one of the following intervals:

    • Overall: any time between start and end date

    • On: in time bucket specified

      • Month 1, Day 4 etc.

    • Since: in time bucket specified or later

      • Month 1 or later, Day 4 or later etc.

Time Windows for Cohorts

  • Overall: for the defined date range of the analysis or the static date range of the Cohort Sample

  • Current: a shortcut for Rolling backward 0 days

  • Rolling backward: Subjects are selected as long as they satisfied the Cohort conditions in the last X days for any of the dates in the date range

  • Rolling backward between: Subjects are selected as long as they satisfied the Cohort conditions between X and Y days before any of the dates in the date range

  • Rolling forward: opposite of `Rolling backward`

  • Rolling forward between: opposite of `Rolling backward between`

Creating a Cohort

In Kubit you're able to build Cohorts in two main ways:

Define Cohorts within the Cohort Builder

At times logic can be specific to a number of conditions or complex and requires an AND or OR condition to retrieve the correct subjects.

When should I use the Cohort Builder?

  • You have specific criteria that you can build the Cohort off of

  • You have to create a Cohort with AND/OR logic to be exported or used on another report for further analysis.

    • Example: All User IDs that streamed something an item AND enabled notifications

Define Cohort from Report Results

More common is to create or inspect a cohort from results within a Report (i.e. Query, Funnel, Retention) to isolate that specific subject population.

To create these cohorts:

  1. Select the data point you want to create the cohort from

  2. Right click and select "Create Cohort" or "Inspect Users"

    1. Create Cohort will prompt you to Save this Cohort for reuse, if you simply want to investigate the users you can cancel this save and view the cohort definition.

Fixed Date Cohort from Report

When creating a cohort from some report types you potentially could be talking about Subjects from one point in time and activity in another point in time.

Here is an example:

  • User IDs who performed a Retention start event of "Sign Up" in May 2024, how many of those users were retained through Month 2 (July 2024)?

In this example Users had to have performed "Sign Up" in May 2024, this is your Cohort Date, and you want that group of users to be analyzed for activity into July 2024. In Kubit we allow you to do this by setting a fixed Cohort Date to ensure accurate results.

If you choose to remove the fixed dates and want to inherit the dates from the report you use this Cohort with click the "x" next to the date range.

Saving Cohorts

Once you've created a Cohort and want to save the logic as a Named Cohort you have two options.

Save a Cohort from the Cohort Builder

If you're building your Cohort from the Cohort Builder, once you've clicked Inspect you will see a "Save Cohort" button in the top right corner of the resulting table. Click this and name your Cohort, add a description and then Save it.

Percentile in Cohorts

When creating a Measure or Query Cohort you are able to limit the Cohort to a percentile.

This will allow you the ability to segment high or low value Subjects within your Cohort.

  • Top 5% of Users performing the Purchase event for items > $100

Compare Cohorts

When creating Cohorts of users it often becomes valuable to understand what events users from one Cohort are doing more or less frequently than another Cohort. Kubit has created a way to do this directly in the Cohort builder. We will compare these two cohorts by using Average Events per User.

When building a Cohort and you want to compare it against another, simply add another Cohort to your analysis, select the date range you want to Compare and use the dropdown to select "Compare".

Cohort Compare Results

The resulting chart will give you the following information:

  • Top 100 or Bottom 100 Events for the first Cohort, we call it "Cohort A". This is determined by sorting the Ratio of Avg Events per User between the two Cohorts.

  • The name of the events in the Top or Bottom events per user.

  • Cohort A information and the Avg Events per User for that Cohort.

  • Cohort B information and the Avg Events per User for that Cohort.

  • The Ratio of Cohort A vs. Cohort B with smart coloring to quickly show events Cohort A performs more often (Green) or less often (Red) than Cohort B.

Inspect User

Let's say you are trying to troubleshoot error scenarios. You already figured out there is a user journey which ends in an error and you found that using Path. But you need another level of detail in order to reproduce the issue. You can now put all the users (or whichever Subject you've selected) who hit this error scenario in a Cohort and see the actual event data for each user. This can help you narrow than what exactly went wrong so you can fix it.

You will define your Cohort criteria and click Inspect:

Then, you can click on any of the user ids in the resulting to see the event data associated with that subject's id.

All data in the Inspect window will be sorted descending by Event Timestamp (the first column). You are able to change the sort of this or any column by clicking the column header.

How to Leverage Cohorts in Other Charts

Cohorts can be used in any Query, Funnel, Retention or Path as a Filter or Segment Breakdown.

Wherever a Segment is applied, customers can also choose from a list of Cohorts.

  • Similarly to dimension values in Segment, each cohort is a “slice” in the returned data structure. In the UI, a Cohort is displayed in the same way as a group in a Breakdown when displaying the data in a chart.

Did this answer your question?