Being able to predict which customers are at high risk of churning can be challenging enough for subscription based businesses such as TV service providers or SaaS (Software-as- a-Service) companies, but the problem is even greater for non-subscription based businesses such as online retailers, restaurants, and free online media sites. This is due to the fact that there are no explicit sign-up and cancellation events as well as no expected stable Monthly Recurring Revenue (MRR). So how does one define a customer churn event without this information? In this blog post, I explore this topic and propose a general approach.
There’s no substitute for loyalty
As crime novelist James Lee Burke famously said, “There’s no substitute for loyalty”. This definitely rings true in the business world as the cost of retaining a customer is much less than that of acquiring a new one. Furthermore, new customers are much more likely to churn than long standing ones, making them less attractive to businesses than long-term customers. Under a subscription based business model, a certain degree of loyalty can be inferred from the explicit action of signing up for a regular monthly service. In non-subscription based businesses, the first step to predicting actionable customer churn is to clearly define the purchase behaviour of a loyal customer. For instance, if a customer purchases a product or service from a business once and never makes a repeat purchase, is it fair to describe that customer as having churned? What actions could have been taken to turn this customer into one of longstanding high value? For example, it is not uncommon for airlines to serve many one-time customers who may not travel regularly on their routes. Converting these customers into repeat customers is a much greater challenge (and much less lucrative in terms of generated revenue) than retaining their existing repeat customers. Loyalty under a non-subscription based business model can therefore be defined through observation of a statistically significant recurring purchase behaviour in individual customers. This can be achieved through, but not limited to, RFM (Recency, Frequency, and Monetary) models, time series analysis, Bayesian models, and/or regression techniques.
Once the loyal customer segment has been identified, it now becomes necessary to quantify the expected regular behaviour of the individual loyal customers. This establishes a baseline relative to which we can later identify churn events. It is important to determine this benchmark as churn isn’t necessarily just a binary event where the customer completely stops all purchases with the business. Indeed, revenue churn at the individual customer level is an important concept to recognize. It is critical to identify a formerly loyal customer that solely purchased from a single business in the past, but that is now distributing their spend across this same business and its competitors. Establishing what this steady state status quo purchase behaviour is can be realized through many of the same algorithms used to determine loyalty. In addition concepts used to calculate Customer Lifetime Value can be leveraged.