Databricks Speeds ERGO Hestia Pricing

ERGO Hestia, a major Polish insurer, has significantly accelerated its real-time pricing capabilities by consolidating its operations onto the Databricks Lakehouse platform, leveraging Databricks Lakebase Mosaic AI Model Serving. This move brings data, features, and AI models into a single, unified environment, enabling millisecond pricing decisions.

Previously, ERGO Hestia’s architecture involved exporting processed data from Databricks to an external PostgreSQL database, managed by a custom adapter layer. This multi-hop approach created extraction overhead and fragmented governance, hindering innovation in a regulated sector.

The Challenge: Scaling Complexity

The insurer operates a sophisticated pricing platform supporting over 100 models and 1,000 variables. While capable, the existing system struggled to keep pace with the demand for continuous model updates and instant customer responsiveness required for real-time B2C pricing.

Operational complexity mounted with the maintenance of external databases and adapter layers, leading to significant overhead. Fragmented governance across systems made lineage tracking difficult, impacting compliance and auditability.

Model deployment velocity suffered due to technical dependencies on synchronized deployment windows and external infrastructure. Data freshness became a bottleneck, with large ingestions causing latency spikes and restricting refreshes to scheduled batch windows.

The Solution: Lakehouse Consolidation

ERGO Hestia’s transformation centers on three key pillars that unify operations within the Databricks Lakehouse:

• Lakebase for Unified Data Serving: Utilizing Sync Tables, Lakebase provides a transactional layer directly on Delta tables, enabling continuous, automatic synchronization. This eliminates manual orchestration and external extraction jobs, establishing a single source of truth for pricing data within the lakehouse.
• Model Serving Endpoints for Direct API Access: These endpoints expose models directly to the pricing engine application, bypassing intermediate layers. This consolidates query execution and data serving into a single managed layer, simplifying the architecture and reducing latency.
• High-Velocity Ecosystem Integration: Models logged in MLflow and registered in Unity Catalog are exposed via dedicated Model Serving Endpoints. This provides a governed plane for pricing experts to validate models against live data in real-time, supporting A/B and regression testing within the Databricks ecosystem.

Existing ETL pipelines required no changes, with data now synchronizing to Lakebase instead of extracting to PostgreSQL. Models registered in MLflow and published to Unity Catalog are readily served.

Incremental Migration for De-Risking

ERGO Hestia adopted a staged approach, starting with low-criticality endpoints to validate performance and stability before migrating mission-critical systems, ensuring uninterrupted pricing operations throughout.

A proof-of-concept phase demonstrated Lakebase’s ability to handle peak request volumes effortlessly, achieving 20ms latency and under 5% CPU utilization under high load. This early success mitigated architectural risk.

Production migration followed, starting with less critical endpoints. Each successful migration delivered consistent data, stable performance, and reduced custom components compared to the legacy architecture.

Traffic splitting at production scale allowed real-world validation, routing half the customer base through the new system. This confirmed the new models handled production load as designed, delivering consistent quote generation and meeting latency expectations.

Business Outcomes: Innovation Velocity

Pricing experts can now deploy models directly within Databricks, automatically synchronized with production data. Serving models via Mosaic AI alongside the Lakebase Online Feature Store eliminates latency associated with syncing external model outputs with live data.

This unified approach accelerates model time-to-market, enabling the pricing team to respond instantly to market conditions. Enhanced governance via Unity Catalog ensures full traceability and auditability for every decision, turning pricing from an IT bottleneck into a strategic growth engine.