Polkomtel S.A.

Challenge:

A special characteristic of how telecommunication companies operate is that they collect  enormous amounts of data which describes the customers and their operations (phone calls, text messages, etc.) This data can later be used to predict the customers’ tendency to resign from using the services or products of a given company (Churn), stop paying, or commit a fraud.

However, the data describing customer behavior changes with respect to amount and type during the whole lifecycle of the customer. This is the major difficulty in utilizing the full potential of this type of information.

 

Solution:

To meet the goals of the project StatConsulting proposed an innovative approach to the construction of analytical models. The main idea of this approach was to use model parameters which would change during the life cycle of customers.

In the project Bayesian models were used to explain the changes of the set of explanatory information as well as the shape of the dependencies between explanatory variables and customer tendencies.

 

Benefits:

The project developed by StatConsulting had the following consequences: 

  • a solution was created which satisfied the requirements of the contracting party,
  • in comparison to competing methods it was possible to obtain better results in areas like prediction effectiveness and the possibility of monitoring.

The Bayesian model approach is capable of determining the behavior of customers during their complete life cycle and shows how the influence of the characteristics describing the customer changes over time.

The development and conclusions of the research project were presented at a prestigious Credit Scoring and Credit Control conference in Edinburgh.