Understand this story
On November 17, 2026, Databricks released an open-source Solution Accelerator for building a real-time next-best-offer recommender for retail banks. The solution uses Feature Views to unify batch and streaming feature definitions, with a p95 latency of 34 milliseconds; it uses a LightGBM ranking model to score all 40 offers directly at the Model Serving endpoint, and during training it uses the create_training_set API to perform point-in-time correct data retrieval to avoid future data leakage. The accelerator is delivered as runnable notebooks, split into two parts: a batch feature version and an extended version that adds streaming session intent. There is currently no information on subsequent developments.
AI-generated from public reports
Latest development · Report
Building a real-time next-best-offer recommendation system with Databricks Feature ViewsDevelopments
1 confirmed developments
Databricks releases real-time NBO recommender Solution Accelerator
Source:Databricks Blog
Reports and discussion timeline
Browse reports and discussion by time. Their evidence levels remain distinct.
Nov 17
Databricks BlogFirst-party
Building a real-time next-best-offer recommendation system with Databricks Feature Views
Databricks has released an open-source Solution Accelerator that uses Feature Views to unify batch and streaming feature definitions and build a real-time next-best-offer recommender for retail banking, with p95 latency of 34 ms. The solution uses a LightGBM ranking model to score all 40 offers directly at the Model Serving endpoint, and during training it performs point-in-time correct data retrieval through the create_training_set API to avoid future data leakage. The accelerator is delivered as runnable notebooks in two parts: a batch feature version and an extended version that adds streaming session intent.