We have developed a Terminus7 machine learning module focus on predicting churn rate based on client behaviour patterns.
Our solution is based on the hypothesis that similar user behaviours implies similar probability to leave the service. The behaviour is based on both demographic characteristics and customer purchase/usage patterns. Therefore, the solution link customers based on the past behaviours. Timeline should be defined in both sides: how far to go in the past data to predict future customer behaviour; how far to go in the future as a prediction borderline with a real business value.
This is a similar approach to Collaborative Filtering techniques used in Machine Learning personalization for e-commerce recommendation systems.
We trained T7 Machine with past customers data and churn rates in order to correlate both information sources with Machine Learning techniques.
After Terminus7 training, Terminus7 demonstrates to be able to predict churn rate with high accuracy and low false positives. Our experience demonstrates that the final Machine Learning prediction quality strongly depends on data quality and volume.