Behavioral Cohort

A behavioral cohort is a group of users defined by actions they did or did not perform in a given period, rather than by who they are or when they arrived.

Membership is the result of a query over the event stream: an event, a counting operator, a time window and optional property conditions. Such a group is either dynamic, recomputed so users enter and leave as their behaviour changes, or static, frozen at the moment it was built. The important difference from an acquisition cohort is that members share no starting point, so plotting retention for a behavioral cohort requires choosing an anchor date explicitly, otherwise the columns of the report do not mean the same thing for every member. The common trap is defining the group on an action that sits next to the outcome: a cohort of people who already purchased twice will convert better, and that is a selection effect rather than a lever.