Empowering Business Decisions with Scalable Churn Prediction

The client, a large technology company, struggled with customer retention and churn prediction. Broad-stroke approaches to customer retention are often cost ineffective and can prompt churn in customers that would not have otherwise left. In the B2B sector, underlying data is often large, inconsistent, and difficult to manage without integrating business knowledge with technical expertise. Left unaddressed, this leads to ineffective churn modeling. The client turned to us for a scalable and actionable churn model to enable more informed retention efforts.

The Aimpoint Digital team worked closely with the client’s business and technical teams to design an end-to-end machine learning pipeline. This pipeline scored customers based on their likelihood to churn, enabling the client to create targeted retention plans and prioritize outreach. The full solution was developed and deployed in the client’s Databricks environment over 24 weeks.

The foundation of any model should be a strong understanding of the underlying data. With churn modeling, this means having a deep understanding of customer behaviors—how they engage with the business, why they leave, and what differentiates positive from negative experiences. Working closely with the client’s business teams, we consolidated insights from over 10 disparate teams and worked closely with the data owners to transform this information into validated features. The resulting feature set created a unified, holistic view of each customer’s life with the client.

The full pipeline, designed using Databricks Asset Bundles and jobs, was created as a robust testing framework. Through a combination of optimizing for Databricks compute, developing custom pipelines, and utilizing MLflow tracking, we reduced iterative runtimes from data to hours, allowing the team to understand the impact of newly developed features on the model on the same day.