AWS has developed a next-best-product (NBP) recommendation system for banking, leveraging Amazon SageMaker and PyTorch to deliver explainable insights from vast customer datasets. This system helps financial institutions predict which product a customer might need next, using deep learning to analyze transaction histories, product ownership, demographics, and behavioral patterns. The solution aims to address the challenges of traditional methods that often fail to capture complex customer behavior over time.
The architecture employs a multi-tower neural network with four specialized towers, each processing different aspects of customer data. These towers are combined using a learned attention mechanism, enabling both high accuracy and per-customer explainability. The system integrates AWS services such as SageMaker AI, Amazon S3, AWS Glue, and CloudWatch to streamline the transition from research to production. It is designed as an architectural overview rather than a deployment guide, offering insights applicable to other domains with heterogeneous data.
The solution requires an AWS account with specific permissions for SageMaker AI, S3, Glue, and CloudWatch, along with Python 3.11+ and PyTorch 2.9+. It also involves setting up IAM roles and policies with least-privilege access. Deploying the system may incur costs for SageMaker training jobs, S3 storage, and Glue jobs, with cleanup instructions provided to avoid ongoing charges.
Source: awsml
