Project Objective
To develop a 'Churn Probability' scoring model and an accompanying DSS to enable the client to facilitate recovery of bills from most likely defaulters.

Client
One of the largest telecom companies in India with annual revenues of $110 mn. The company has subscriber base of more than 2.6 million spread across 209 cities.

Approach
Each month's data comprised of more than 150 variables and 65,500 records, wherein each record represents an individual customer. We used PROC LOGISTICS of SAS to identify variables that have no relevance to the probability of churning. Thereafter, various methodologies were considered for building the model.

Neural Networks Decision Tree Logit Output is required as a Probability number - A Continuous variable Decision Tree Method for Model Generation   Output is always a Discrete variable   Output is Continuous variable   Output is always a Discrete variable   Better at predicting Binary or Discrete variable   Better at predicting Continuous variable   Better at predicting Discrete variable

Solution and Benefits
We built a DSS system to drive better decision-making. The DSS allows managers to control and analyze using a 'What if' simulator, the return on investments in churn management. The DSS also provides a comprehensive comparative assessment of customer profitability as measured by 'Average Revenue Per User' (ARPU) vis-à-vis customer loyalty indicated by the 'Tenure Base' of the customer on its network.