Understanding fixed line broadband churn in MTNL using limited data

Abstract

Broadband is an important service for the legacy landline operators such as MTNL and BSNL. Firstly, it arrests landline churn (to mobile) and secondly, increases revenues, being a higher ARPU service (to mobile). With the growth of new Tele-Co s wire-line network and alternates such as wireless data cards, lot of churn is taking place in fixed line broadband. In this perspective, churn of even one broadband customer, represents loss of significant future cash flows. The cost of retaining existing customer being less than the cost of acquiring new one, it makes lot of sense for Managers to invest in retention efforts. For maximum returns, it is utmost important to spend the allocated (limited) budget on potential churners only. This requires identifying them accurately, preferably with whatever limited data at hand, for taking action promptly. While literature is abundant with mobile churn studies, only a few studies exists in respect of fixed line broadband churn. This study aims to find the variables that differentiate the wire-line broadband churners from the non churners and develop a model for predicting the churners using (limited) data available within the organization. Modelling techniques of Logistic Regression (LOGIT), Classification Tree (CRT) and Artificial Neural Networks (ANN) are used to predict churners and arrive at variables having differentiating power between a potential churner and non-churner. Subscription duration, Month of Joining, No. of faults, Service Area and Subscription to company s other services are found to be important predictor variables. The accuracy of ANN, though slightly more than LOGIT was of similar magnitude. Corroboration with exit interviews of churners suggests limitation in achievable prediction accuracy, due to reasons internal to the churner, not getting reflected in the company s (limited) internal data. Based on the observations of modelling and exit interviews, inferences and recommendations for MTNL are drawn.