Death, Taxes, and Unsubscribes: Modeling Email Churn with Survival Analysisby Zuzia Klyszejko
For many online businesses, email is the primary channel of communication with customers. For marketers, email is appealing because it is a direct line to customers, has a wide range of content flexibility, and, most importantly, does not impact marketing budgets. While sending marketing emails may appear free, the decision to send an email is not free of risk. When customers receive emails and choose the unsubscribe, the resulting churn decreases the long-term potential value of the email list as marketers are forced to turn to costly, less personal channels to reach lost customers. Therefore, incorporating probability of churn into the send decision is imperative to marketers who wish to maximize the long-term value of their email list. Predicting email list churn makes for an interesting classification problem to solve with machine learning tools. In this talk I will focus on churn prediction using logistic regression and survival analysis with custom time-dependent offline validation method to simulate algorithm performance and user behavior.