Introduction

Businesses are always interested in studying churn behaviors among their customers. Understanding churn can identify factors that potentially correlate to customers leaving but can also be used as a predictive force to identify at-risk customers and proactively engage them to prevent churn. There are various methods to model churn, depending on your domain and use-case. This post will explore 3 unique approaches to model churn:

1. RFM Segmentation
2. Classifier — Based Predictions
3. Survival — Based Models

RFM Segmentation

The simplest approach is by grouping customers into segments or “personas”. The approach is simple in that it simply uses 3 features: Recency, Frequency, and Monetary value. These terms, used most often in marketing, are roughly defined as:

This last dimension aims to identify how meaningful or valuable were those returned engagements/visits? If the unit of metric is purchases, then the monetary value can simply be the customers average purchase price.

You can use purchase or engagement as the unit of action depending if your business model is based on customers returning to purchase (e.g., e-commerce, SaaS B2B) or customers returning to further engagement (e.g., Instagram, twitter). Measuring the monetary value for engagement may require some prior weighting of types of engagement (i.e., uploading an image is perhaps _more _meaningful than simply logging in and scrolling through a feed).

#churn #customer-success #data-science #customer-churn #data analysis