E.ON: Retention Campaign Cut Contract Terminations by 50%

Identifying the Right Customers for Retention Efforts

To enable timely customer engagement, we designed and automated a process running on Microsoft Azure. Our tasks included:

1. Designing an analytical datamart containing all relevant data needed to develop a predictive machine learning (ML) model.
2. Automating the transformation and loading of data from the data warehouse into the datamart.
3. Using the datamart on the Databricks platform to train, evaluate, interpret, and automate an ML model aimed at:

• Identifying customers at risk of switching to a competitor.
• Determining the key factors driving customer dissatisfaction, enabling E.ON to proactively target these customers with the most suitable retention offers.

From 300 Core Attributes to Over 7,500 Advanced Predictors

As part of the business analysis, we mapped out the key aspects of customer churn, contract terminations, supplier switching processes, and the regulatory framework governing these activities. We defined the data requirements not only for the development of the ML model but also for future use cases.

Working closely with E.ON’s business stakeholders, we identified critical data within the data warehouse, designed the necessary transformations, and built the concept for the analytical datamart. We then synchronized the automated transfer of data from the datamart to the Databricks platform.

On Databricks, we expanded approximately 300 core attributes into a rich set of over 7,500 advanced predictors. The ML model was developed and trained on an analytical dataset comprising 10 million records and 750 attributes.

The model’s output—a churn probability score for each customer—is regularly exported from Databricks back to the data warehouse, enabling E.ON to proactively engage with at-risk customers. The entire process is fully automated and refreshed monthly.

ML Model Identifies Twice as Many At-Risk Customers Compared to the Original Expert Approach

Using the machine learning model to select customers for E.ON’s retention campaign, we achieved twice the accuracy in identifying at-risk customers compared to the previous manual, expert-rule-based method—and 21 times higher accuracy than random targeting.

The proactive retention campaign based on the model’s results reduced contract terminations by 50% among the top 5% of the most at-risk customers.

Key Results:

• 50% reduction in contract terminations among the top 5% most at-risk customers through a proactive retention campaign driven by the ML model.
• 2x higher success rate in identifying at-risk customers compared to the original expert-based selection.
• 21x greater targeting efficiency compared to random selection.