The result is a predictable decay of a given set of customers over time, in the same way that uranium isotopes decay. There are three statistical analysis perspectives on this customer list decay.
-
You can focus on a set time period for decay, say 6 months, and ask “what is the probability that a customer acquired 6 months ago will leave in the next month.” You could train a static statistical model on the basis of training data, where each customer is classified 0/1 as to whether they left in the 7th month.
-
You can use basic survival analysis to understand the general survival rate over time of a typical customer, to use as a parameter in business planning. This would yield a function that you could use to determine risk for the typical customer at any given time.
-
You can add predictor variables to the survival model to model churn risk not just for a typical customer, but for a customer who fits a particular profile. This would allow more targeted interventions for a given customer to lower the probability of churning.