Case Study: Forecasting Subscriber Attrition

A major mobile phone service provider was anxious to move from reacting to sudden spikes in customer attrition to managing customer attrition proactively. Data Miners developed a churn forecasting system that predicts the number of active subscribers there will be on a daily basis for a year or more into the future. Unlike aggregate level forecasts based on historic churn rates, the system developed by Data Miners uses the specific attributes of each customer segments (or even individual customers) to calculate survival curves. Many variables affect the expected lifetime of a customer: credit class, rate plan, acquisition channel, calling behavior, payment behavior, to name a few. At any given time, the active subscriber base contains a mix of customers with different values for these explanatory variables. Our forecast takes all these factors into account along with the characteristics of newly acquired customers and the expected characteristics of planned future additions to the customer base. One consequence is that the forecasting tool is also useful for "what if" analysis. Managers can use the tool to forecast the future effects of alternate customer acquisition strategies that might be adopted.

The chart shows the survival curves associated with customers from four different acquisition channels. All the curves show an acceleration in cancellations at a tenure of one year when subscribers typically come off contract. The effect is much stronger in the red line that represents customers acquired through independent dealers. Hence, the company could reduce "anniversary churn" by reducing its reliance on this channel.

Since building our first survival-based forecast for this mobile phone service provider, we have built several more for telephone companies, internet service providers, newspapers, and other subscription-based businesses.