How Cohorts and Buckets are Formulated

When setting the parameters for a retention analysis, you will select your desired number of cohorts and buckets, as well as an End Date which limits the time frame of your analysis.

However, it’s important to note that selecting an End Date also determines the way your cohorts and buckets are defined. In particular, the analysis time frame will stretch back and orientate itself around the desired number of cohorts and buckets – whichever is greater.

The easiest way to illustrate this is through a couple of examples.

Example A

Let’s say you set the following parameters:

• 3 monthly cohorts
• 10 monthly buckets
• End Date = 1/1/2023

This means you will analyze 10 months of data (a.k.a. buckets) leading up to the End Date – dating back to March 1, 2022.

The three cohorts will consist of users who had a Starting Point during each of the first 3 months of the time frame (March, April and May).

This table illustrates how the results will be distributed:

As is always the case, the “Month 0” bucket refers to a cohort’s first week/day/month of activity.

“N/A” means there is no data for that bucket because the analysis will have already come to an end. For example, Cohort #3 would only have data for Months 0-7, since Months 8-10 would occur after the End Date of January 1, 2023.

Example B

Now imagine you set the following parameters:

• 10 monthly cohorts
• 3 monthly buckets
• End Date = 1/1/2023

In this case, you will analyze 10 monthly cohorts leading up to the End Date. As a result, Cohort #1” will consist of users who had a Starting Point during March 2022.

Each cohort’s behavior will be analyzed across 3 monthly buckets, starting from the month of their Starting Point (“Month 0”). This table illustrates how the results will be distributed:

Again, the “Month 0” bucket refers to a cohort’s first week of activity.

You would not have data for certain buckets (represented as “N/A”) because the analysis will have already come to an end. For example, Cohort #10 will only have data for the bucket “Month 0” because the next two buckets would occur after the End Date of January 1, 2023.