Churn Prediction
In the complex telecommunications scenario in Spain, Vodafone is faced with the constant challenge of not only attracting but also keeping its customers. In the face of intense competition, it is crucial to identify those customers most at risk of migrating to other companies. Traditional churn prediction models have often focused only on identifying symptoms of customer dissatisfaction. However, Vodafone sought to go one step further, seeking not only to predict who would leave but to deeply understand the causes behind these decisions.
At WhiteBox, we have developed a tailor-made causal inference framework for Vodafone, taking advantage of open-source tools such as Causal ML and CausalNex. This approach has allowed us not only to predict the likelihood of customer churn but also to decipher the forces behind that trend. To implement this model, we used MLFlow for efficient management of the model lifecycle and Apache Airflow for task orchestration, all integrated into the robust infrastructure of Google Cloud Platform.
This advanced model has transformed Vodafone's approach to churn prediction, allowing Vodafone not only to predict which customers are at risk and with what probability but also understand the reasons behind their possible departure. In addition, this model predicts how specific retention actions (such as promotions and targeted offers) can influence the customer's final decision, thus optimizing the impact of those strategies. The result is a more accurate and effective application of retention measures, aimed at those customers that would benefit the most from them and avoiding those who would cancel anyway.
- Development of models with more than 5,000 customer characteristics.
- Evaluation of the churn risk of more than 5 million monthly customers.
