Retaining or losing customers as they “churn” — Professor devises new competing risk model

MIT Sloan Assoc. Prof. Michael Braun

Businesses often spend a lot of money trying to retain customers. Many direct their retention activities toward all of their customers and hope that enough respond for the effort to pay off. But some customers leave anyway.  A more effective approach recognizes that customers are different and their likelihood of departing—a phenomenon known as churn in the business world—varies among individuals and over time.

Some customers churn for reasons a business can control. These customers may be unhappy with the price or product, or they may prefer a competitor. Others churn because they move away or die or go bankrupt—matters a business can’t control. An efficient strategy targets those customers likely to churn for controllable reasons and does not overspend on customers likely to leave for uncontrollable reasons. And when evaluating the success or failure of retention marketing activities, managers should take the incidence of uncontrollable churn into account.

With my colleague, David A. Schweidel of the University of Wisconsin, I devised a mathematical model that helps to resolve both the targeting and measurement problems. The model generates probabilities for when customers will churn. With the model and data on a company’s past customer retention, a business can determine which customers to target and what the payback is likely to be. The model can be used with any subscription or membership-based service, such as cable television, a health club, or direct-debit charitable donations.

The model we created is known as a competing risk model, a statistical tool used commonly in medical research. Doctors want to know how a treatment for a disease affects the lifespan of a patient, who can die of any number of other causes, but can only die once. By taking into account when and why other patients died and which types of patients survived, a competing risk model can untangle the interdependent complexities of disease and treatment.

Similarly, our model uses information about past customers to identify which customers today are more likely to churn and when. It takes into account the fact that some reasons for churn are uncontrollable. For example, if it turns out that uncontrollable reasons for churn happen frequently, the return on investment from reducing controllable churn is lower, because customers are more likely to churn from other reasons earlier on.  And customers who have been with the firm a long time have lower intrinsic propensities to churn, because if they didn’t, they would have churned already.  With our model, a business can estimate the payback, in terms of discounted cash flow, from its customer retention activities, and it will have a better idea of which customers to try to keep and which ones to let go.

Read more in Marketing Science 

Michael Braun is Associate Professor of Marketing 

Share your thoughts

Post a Comment

Required fields are marked *

*
*