Improving subscriber retention is one of the best-understood problems in the domain of business intelligence, but it is typically undertaken based purely on CRM data and billing history. While these can provide reliable indicators of potential churn risk, we have found that fusing in behavioral data, particularly consumption information can significantly increase the accuracy of churn models.
We will typically run an initial segmentation of the subscriber base using the CRM variables to identify those who are most valuable and then run different analyses on these. A good indicator of loyalty to a TV service is the level of engagement in the service. This is not merely a matter of how frequently they used the service, but also how much they use it when they do use it.
The chart below shows a heat-map of every subscriber of one Pay TV service. The x-axis shows how frequently the platform is used - with the right-hand edge being subscribers who use it daily. The y-axis is how many hours a day it is used. The brighter the color on the heat map the more subscribers there are who engage in this manner.
We can see for this platform, the vast majority are using the service each day for many hours, but there remains a significant portion who are using it fewer hours and less regularly. These are the subscribers who are most likely to churn in the future and the ones that require special care.
We created a decision tree using a random forest model for these subscribers and identified three key variables that were the most reliable indicators of churn for this segment.
The most robust indicator was a preference for football. There is a lot of football on this MVPD and was a key driver for a subscription. Heavy football viewers, regardless of other habits were unlikely to churn. Beyond that, the two other elements were the usage of DVR and OTT services.
We worked with the MVPD to create services specifically to target the at-risk segment with DVR and OTT offers that resulted in a significant increase in retention amongst that group, contributing to an overall 2% reduction in churn across the entire MVPD.