Cohort is a group of players defined by their entry moment, usually the week or month of first deposit, tracked over time.
Cohort analysis answers the question averages cannot: did the change we made improve anything? Comparing "this month's retention" with "last month's" mixes players who joined yesterday with players from two years ago; comparing the March cohort curve with the April one isolates the effect.
Practical cautions:
- Segment by channel. Affiliate, paid media and organic cohorts have different curves; the average across them describes none of them.
- Freeze the definition. If the "active" rule changes, the whole historical series must be recalculated, otherwise you are comparing different things under the same name.
- Store the definition with the number. A dashboard carrying the rule that produced the cut avoids much of the meeting debate.
Updated
All terms