What about Cohorts Analysis?
In our past article of this series we have covered first example of using Historical CLV, Average Revenue Per User (ARPU) and have realized its pitfalls when dealing with large customers base (both existing and perspective ones). Now it is time for an alternative method (more common among marketers) that relies instead on the concept of cohorts (defined as a group of customers/clients/subscribers who share a set of attributes; in my example today, a cohort is represented by those who joined or made their first valid purchase in a given month). As a consequence, instead of computing overall average monthly revenue per customer, this method will calculate metrics across time of each cohort.
With reference to our earlier company XY case study of Alex and Rob as total customers, we can easily state that our January cohort will consist of Alex and our May one will include Rob only. Using same transactions table as shown here, we have that a simple version of company XY cohort analysis would look like this below:
We used as main metric Total Revenue per customer from transactions and “plotted” along the vertical-axis/time of each cohorts (i.e. Jan-13 is Alex’s cohort and will consist of all his revenue going forward placed in the relevant rows since his first purchase month – for instance the Jan-13/150days cell highlighted in blue above will include his May 15th transaction of $5). Moreover, as you can see in below chart we can use instead Cumulative Revenue per customer as metric for the cohorts, with a very easy representation of how much your cohort average spending over a time period would look like.
Cohort analysis is fairly easy to prepare and factors in the fact that not all customer months are the same, thus enabling us with an interesting view of variations across lifetime of a customer. Or instead, assuming that new customers can be same across different months, in our example above it is possible to expect that our May customers will have a no-purchases period of 3-4 months and then will come back to purchase again as the January cohort did.
As cohort analysis can give us a more transparent picture of CLV, at the same time it can be inaccurate if the business (or the market) is frequently changing. Factors such as seasonality, last minute promotions, new entrants in the market or totally different product assortment year by year, will result in new cohort months to have very few things in common to your first/older cohorts base, thus making our existing data a less reliable predictor of future performance. And finally, as cohort analysis technique is based on actual m/m changes, it will require a somehow long time (let’s say at least 18months of business) to get full cohort picture and assign a 18-months CLV for your company.
In next article we will discuss more in depth some use cases of cohorts analysis and CLV (i.e payback period and repurchases rate).