What is Cohort Analysis?
First, the broad definition: a cohort is a group of individuals linked together, typically by age. It’s basically the same thing as a generation.
In practice, though, a cohort is any group moving through a system together.
For games, you can think of a cohort in terms when they started. If your game or app started on January 1, you have a starting cohort that began then. You decide when that group definition ends, and the next one begins. You might pick the first week of service, the first month, the first year, the time until the first content patch--whatever makes sense in thinking of that group as different in some way than the next.
Cohort analysis is therefore just a fancy way of saying you’re looking at the people who started playing your game or app at different times.
Cohorts Pass Through Phases
Just as we think about a generation going through various stages of human development, cohorts have stages as well.
For example, imagine a player coming to a game via some advertisement. They visit, sign up, try the tutorial, make it to level 1, spend money, etc. Second, imagine a player coming to join a more social community.
The genius game designer Amy Jo Kim suggests a series of stages for most members of an online community: Visitor, Novice, Regular, Leader, Elder.
Whatever stages and phases you think make sense, the important part of cohort analysis is that different cohorts are going to be in different stages. It sounds obvious, but it’s easy to forget that your newbie player is experiencing very different things than your long-timer. The wrinkle is that that same set of stages might actually play out differently for different cohorts.
Cohorts Have an End
Also just as with generations, cohorts don’t last forever. They go through the life cycle and then they go away. In real life, that’s death. In games and apps, that’s churn. And again, different cohorts may have different lifespans.
Looking at Cohort Data Independently
There are several good reasons to look at each cohort independently. This means that whatever tool you are using for your analytics should be easily able to create a group based on the variable “start date” and give you a choice for your cohort’s date range, e.g. Jan. 1-31.
In our Katana Analytics Engine, we consider this to be just another form of segmentation. You might think about segmenting a group by, say, gender to compare men vs. women. Think the same way for segmenting by “start date,” and you can compare the January group vs. the February one, etc.
Use Cohorts to Determine New Player Aquisition
Most game developers will look at Daily Active Users (DAU) or Monthly Active Users (MAU) to help get an idea of new player aquisition. But you can also look at cohort data to get a different take on the same question.
For example, let's compare your first three month-long cohorts. If the game started Jan. 1, then you have a January cohort, a February cohort and a March cohort. Let's say you find that the January cohort had 60k starting players in it, February had 50k, and March had 40k. This tells you you're not onboarding as many players as you used to.
Or say that Cohorts A, B and C all have 50k new players, but do they spend or convert or churn the same? Maybe Cohort C has an Average Revenue per User (ARPU) of $4, whereas A and B have an ARPU of $2.50. That suggests that whatever drives people to spend is working much better now than it was two months before. Or, it may be that Cohort C has much better social value than the other two, i.e. the players in it are getting each other to play and spend more and their community is simply better.
Looking at Cohorts' Age and Period
Now let's add two more pieces, age and period. In game analytics 'age' is just how long the player has been playing. Some cohorts are older than others. And, by virtue of that age, those players will be at a different point in the life-cyle of the game.
'Period' is when on the real-world calendar things take place. Without period you can't understand big real-world events, e.g. play is dropping off in September among my young player, so what's wrong with my game (Answer: Nothing, it's just the school year).
So let's say you notice that the B cohort is churning out faster and you can’t figure out why. Then you realize that the developers put out a new patch in April, and you start to think about where each cohort is in its age (life cycle stage). You ask your developers what was in the April patch and they tell you “Oh, we changed the way people do quests in the middle levels.” Bingo.
Now you put the various elements (age, period and cohort) together, and you realize that it takes two months to get to the middle levels. Therefore Cohort A is already past that stage, Cohort B is right in it, and Cohort C is not there yet. You’ve disentangled a “period” effect from age and cohort. More practically, you can now go back to the developers and let them know that the middle-level quest content change appears to be churning out your players. If you hadn’t looked at it by cohort, you never would have seen the problem.
By the way, I have to give a shout out to a great little book that I refer to on occasion - Cohort Analysis by Norval Glenn. The author was a real ninja of cohort analysis and much of what we talk about here at Ninja Metrics regarding cohort analysis comes from his work.