Cohort analysis takes the ARPU approach one step further. Instead of calculating an overall average monthly revenue per user, cohort analysis calculates an ARPU per month per cohort (a cohort is a group of customers who share an attribute or set of attributes; in this case, a cohort is defined as those who joined or made their first purchase in a particular month).
Looking back at our previous example, we can calculate monthly ARPUs for our January cohort (consisting of Alice) and our May cohort (consisting of Bob).
|Cohort||Month 1||Month 2||Month 3||Month 4||Month 5||Month 6|
Cohort analysis is simple to calculate and recognizes that not all customer months are the same, giving us a nice view of the variation across the lifetime of a customer. If we can reasonably assume that our January customers don't differ greatly from our May customers, then we might expect that our May customers to follow our January customers and go “silent” for a few months.
While cohort analysis gives us a clearer picture of CLV, it can be inaccurate if the company and/or the market changes. It's entirely possible that some change, such as seasonality, new promotions, new competitors or new product categories, will result in our June customers looking very different than our January or even May customers, making our existing data not a very good predictor of the future. Furthermore, since cohort analysis relies on looking at actual month-to-month changes, it does take a long time (two years to be exact!) to get a cohort with enough data to calculate a two year CLV.