Sample Projects and Implementations

The main objective of the project was to assess the bank’s customers in view of their churn risk. Additionally, the project included simulating the effectiveness of an action aimed at retaining the customers at risk of leaving (churn) by offering them the right kind of service.
In the first place a 100-feature profile of several million customers was created. With this purpose demographic data along with information on a customer’s business history were entered into the R2 SQL 2008 table. Then, the data quality was assessed and corrected (e.g. the ratio of customers at risk of leaving was increased) using SSIS. Thus prepared data were later used to classify the customers in respect of their risk of churn with the use of Microsoft Decision Trees algorithm (a version of the classical decision trees algorithm, implanted in SSAS server). The modeling revealed a list of the strongest predictors (attributes directly or indirectly influencing the customer’s decision of discontinuing business). The accuracy and the reliability of the model were tested using the embedded SSAS server functions and by consulting a business analyst. The resulting model turned out to be 65% effective in identifying customers at risk of churn, which  amounted to an  accuracy increase of over 20 times as compared with random guess.


The created model was then employed to assess the risk of particular customer churn. Allocating to particular customers their churn risk helped to choose data for the segmentation model (a model constructed with the use of Microsoft Clustering algorithm), enabling to develop those customers’ profile.
 The data concerning the customers identified by the decision trees model were also used to perform a simulation of a “what would happen, if…” type, aimed at revealing the effectiveness of the attempts to retain the customers in question. The predicted effectiveness of the prevention campaign selected in this way amounted to 25%, which means it would be successful in retaining every fourth customer about to churn.

  1. A list of customers at risk of churning was compiled, along with the estimation of the probability of their churn.
  2. The profile of a customer at risk of churning was identified.
  3. The most effective strategy for retaining customers at risk of churning was determined.