Churn models introduction

Why are “churn models” so important? Let’s give an example. Typically, the yearly churn rate for a telecom operator (i.e. a mobile phone operator ) is 12%. That means that each year, a mobile phone operator loses 12% of its customers (this is usually not a real problem because, in a saturated market, the same telecom company gains 12% of new customer and everything usually stays “in balance”).

If we manage to detect these “churners” and prevent them from leaving, the telecom operator will gain:

(20 €/month x 12) x 1,000,000 x 12% = €28,800,000 euros

In the above equation, we used the following hypothesis:

  • subscription fees of 20 €/month paid out over the course of a year
  • that the small mobile phone operator has about 1,000,000 customers
  • And that the annual churn rate is 12%

Let’s say that we are using TIMi for our predictions. TIMi is able to correctly detect “in advance” 40% of the churners.Then, the telecom opeator can offer to the people who are about to churn “free talk time” (or another incentive) and, as a result, manages to keep 25% of the potential churners. In this example, the value of the ranking generated with TIMi is:

28,800,000€ x 40% x 25 % = 2,880,000 €.

Other classical, best-in-breed, commercial datamining software, is only able to detect “in advance” a maximum of 30% of the churners. The value of their ranking is thus only:

28,800,000€ x 30% x 25% = 2,160,000 €.

Business-Insight works hard on the prediction accuracy of TIMi because of the massive added value each predictive model has for our clients. In this example, nearly a million euros in addition to a standard modelization tool for a single predictive model. The superior accuracy of TIMi is demonstrated by its outstanding performance at the various KDD cups.