Retention of customers is hard (no-contract compared to postpaid customers); no clear definition of churn/churners; short-term nature of prepaid products; customers moving among providers; False identification of churners; Seasonality.
New definitions for “active customer base” and “churn” are required, as well as new features for better modeling of churn.
Targeting of customer sub-groups with optimal combination of predictive value and potential financial benefit, feature engineering that takes into account customer profile and history, product usage patterns, recharge history, recharge type / package selected, voice/data profiles, complaints etc. and advanced ML for identification of highly probable churners.
Deficient Predicting prepaid customers’ churn for telecom operators
DESCRIPTION WHAT WE HAVE DONE/ACHIEVED
Ability to correctly identify thousands of churners with less than 10% false positives. Automated pipeline for data preprocessing, labeling, feature engineering and model training that shares predictions with marketing department to coordinate relevant offers and incentives for retention of identified customers.