ERGO Hestia implements real-time insurance pricing with Lakebase

ERGO Hestia relied on a legacy multi-hop architecture in which data was exported to an external Azure PostgreSQL database, paired with intermediate application layers. This setup created high operational maintenance, latency bottlenecks, and complex extraction jobs – forcing large data updates to occur during restricted off-peak hours to avoid impacting live pricing performance.

They replaced their external database with Databricks Lakebase and Databricks Model Serving Endpoints. By consolidating query execution, data serving and model inference into a single managed layer, they eliminated intermediate caching layers, reduced network hops and kept data governance unified within Unity Catalog.

To ensure zero downtime for critical insurance operations, ERGO Hestia used a "baby-steps" incremental migration strategy: 

• Phase 1 (Proof-of-Concept): Tested small, low-criticality endpoints to measure CPU, memory and throughput metrics under heavy synthetic load. 
• Phase 2 (Production Rollout): Migrated remaining endpoints sequentially from low-criticality to mission-critical systems. 
• Phase 3 (Traffic Splitting): Routed 50% of live user traffic through the legacy architecture and 50% through the new Databricks system to validate real-world performance side-by-side.

• Low Latency: Achieved steady 20 ms response latency during peak load testing. 
• Resource Efficiency: Maintained under 5% CPU utilization under a high stress load of 40 requests per second. 
• Zero Downtime: Successfully completed a 50/50 production traffic split with identical quote generation accuracy and no service disruptions.