A Multi-Model Approach for Predicting Churn Customers using Longitudinal Data

Due to the dynamic nature of the Retail market and increased competition among the major players to attract customers, it is becoming increasingly important to focus on customer retention. Also, attaining a new customer is usually costlier than retention. The first step prior to retaining customers is to first identify potential churn customers. Hence, it’s important to identify the customers who are about to churn and address the pain points. Typically churn modeling is performed on cross-sectional data, and the results provide an overall effect of each variable on the entire data set. This creates a loss of information about the temporal behavior of customers. In this paper, a new framework is proposed to overcome these shortcomings. A multi-model approach is followed to account for both cross-sectional as well as temporal data for predicting the churn probability of a customer. Focus is kept on retaining high-value customers, and probable customer-specific features leading to churn are derived to help the business act more precisely. Customer profiling is integrated with the results to enable the business to make more informed decisions and act at a customer cohort level. A comparative study between the proposed new framework and the traditional RFM model shows a significant improvement in the results using the proposed framework.