Case Study: Neo Financial cuts fraud-stack infrastructure cost 80% with a Materialize online feature store
Key results
The challenge
Neo Financial needed a real-time feature store that could serve fraud-detection features with sub-1-second latency while limiting operational complexity and cost. Its prior approaches, including DIY MongoDB, ClickHouse, Flink, and warehouse-only setups, required heavy engineering maintenance.
The solution
Neo built an online feature store on Materialize, streaming MongoDB data and defining aggregates in SQL. Incrementally maintained views serve fresh features, and new features ship through SQL and dbt workflows.
“Our fraud losses have substantially decreased, and the infrastructure spend for this system has gone down by about 80 percent.”
BRBrandon RochonHead of Data, Neo Financial
The results, in context
Neo reported infrastructure spend for the system down about 80%, sub-1-second P99 feature latency for point-of-sale approvals, and more than 20x faster feature delivery (hours instead of days).