"There’s tremendous value in accurately predicting churn at a customer-by-customer level at telecom companies. If a company offers discounts to people who would have stayed anyway, it has wasted its money. A lack of appropriate targeting can also make it overlook people who might leave for a competitor." "...Telekomunikacja Polska, part of France Telecom-Orange Group and the largest fixed-line provider for voice and broadband services in Poland. The company wanted to quickly find ways to predict and address churn among its customers more effective than the traditional methods, including analysis of declining rates of use and calculations of lifetime customer value based on how long customers stayed with the service and how much they spent.
They built a ''''social graph'' from the call data records of millions of phone calls transiting through its network every month—looking, in particular, at patterns of who calls whom and at what frequency. The tool divides communities into roles such as ''networkers,'' ''bridges,'' ''leaders,'' and ''followers.'' For example, it detects the networkers, who link people together, and the leaders, who have a much greater impact on the network of people around them. That set of relational data gives the telecom-service provider much richer insight into who matters among those who might drop its service and, therefore, how hard to try to keep its most valuable customers. "
"...call data records of millions of phone calls transiting through its network every month—looking, in particular, at patterns of who calls whom and at what frequency."
...the accuracy of churn-prediction model accuracy has improved 47% allowing focus on how hard to retain influential customers.