Predicting Customer Intent at Scale for a Top Sportsbook
Explore how Aimpoint Digital’s Decision Sciences team designed, developed and deployed a state-of-the-art neural network capable of accurately estimating the likelihood a customer would bet on all possible sporting events each day. The solution was integrated into existing loyalty rewards campaigns, driving positive return on investment (ROI) through improved targeting.
The Challenge
The client was looking to become more efficient with their loyalty rewards budget by improving their ability to target which customers were most likely to bet on a particular sporting event. Given the client processes millions of bets across hundreds of sporting events each day, the solution needed to be able to generate predictions for every customer for every event in minutes.
Our Approach
The Aimpoint team was embedded within the client’s product team as both a core developer and advisor to the overall execution of the project. The project was broken down into four key steps, executed over three months:
- Set up an iterative process to work with business analysts and SMEs to identify key data sources, ideate potential features, and review model outputs to drive continued enhancements to the project
- Design a custom neural network architecture using state-of-the-art research that could best exploit patterns in the underlying data to correctly identify customer actions
- Implement a scalable model training and deployment framework using Ray and Databricks GPU clusters to train models on hundreds of millions of records in hours
- Develop robust model evaluation framework to quickly identify areas where the model is not performing well to guide future improvements
This process allowed the team to continuously and quickly solicit feedback from the business and incorporate it into each iteration to further enhance the solution.
Results
The sporting event intent model supported a larger initiative to take advantage of improved targeting to determine which users should get rewards, communications, or nothing. After extensive tests, the model was able to identify profitable segments of customers, reducing costs for event-based campaigns by over 80% when compared to broad-stroke approaches.
Aimpoint developed a distributed compute framework to train large models across many GPUs using Ray and Databricks. This framework empowered the client to rapidly build additional customer intent models for other areas of the business.
By combining SOTA research with business understanding, the intent model was able to both capture recent patterns in behavior and long-term preferences, which resulted in a model that significantly improved predictive accuracy over existing solutions.
Key Takeaways
The Aimpoint Decision Sciences team goes beyond just building a model, we set up processes and frameworks that extend the value of our engagement far beyond the solution. The client was not only able to realize near-term value from leveraging the customer intent model in production to make targeted marketing decisions, but long-term value as well since the project was adapted as a template for all future projects, ensuring technical excellence across their ML engineering and data science team.
Related case studies
Let’s talk data.
We’ll bring the solutions.
Whether you need advanced AI solutions, strategic data expertise, or tailored insights, our team is here to help.