A quick intro to cohort modeling in Causal

What is cohort modeling?

In order to understand your customers at a deeper level, you'll often want to track them on a more granular basis, using time-based cohorts. This is because customers can behave differently at different points in time, for example:
  • Retention and expansion rates often follow a pattern depending on how many months they've been a customer
  • Churn rates may differ depending on what time of year a customer joined
  • Spend or engagement may be higher in initial months after sign-up, then drop off
  • If you have a new onboarding flow, more customers might retain in the 2nd month, (vs. when there was no new onboarding flow)
To forecast on this cohort basis, you need to be able to split your customers into different cohorts, so you can apply the correct assumptions (e.g. churn, retention etc) at different points in time.

Cohorts in Causal

In Causal, Cohorts act as a category that reflects the time period of the model. For example, if a monthly model goes from Jan '22 to Dec '22, adding cohorts would add 12 items, one for each month in the model.
You can access cohorts by explicitly adding the cohort category in the variable or by referencing cohorts in the formula. Below is a simple example of using cohorts ->

Cohort of leads that convert into New Customers

Let's break this formula into parts:
Example Inputs:
  1. 1.
    New sign-ups is 1,000 for our first month and grows at 5%
  2. 2.
    Activation of cohort uses relative time so that 1st month is 45%, 2nd is 25%, etc.
Simply this formula is saying "Signups multiplied by activation rate %"
What is the "cohort" and "t-cohort" doing?
  • By putting cohort as the time modifier of the New sign-ups variable, we are telling Causal to use the new sign-ups for Jan in the Jan cohort, Feb in the Feb cohort, etc.
  • By using t-cohort as the time modifier of the Activations variable, we are telling Causal to use the 1st month activation rate (i.e. 45%) for the first month of the cohort (Jan'22 for the Jan'22 cohort), the second activation rate (25%) for the second month of that cohort (Feb'22 for the Jan'22 cohort), and so on.
Deeper dive on t-cohort
Let's consider the Feb '22 cohort, in Mar '22, in a model that begins in January '22:
  • cohort is 0 for Jan '22, 1 for Feb '22, etc.
  • t is 0 for Jan '22, 1 for Feb '22, 2 for Mar '22 etc
  • For our worked example, t is 2 and cohort is 1. t-cohort will return 1, so I will be applying the 2nd month activation rate (25%). This is correct as Mar'22 is the 2nd month of my Feb'22 cohort.
  • If the month was instead Feb '22, then t-cohort would be 0 (1-1) corresponding to the 1st month activation rate of 45%
Note: t is a helper variable (also known as timestep or date).