Poland's leading insurer ERGO Hestia transformed its pricing platform by consolidating data and model serving within Databricks Lakehouse using Lakebase and Mosaic AI Model Serving. This shift eliminates external systems, slashes deployment times, and enables real-time B2C pricing updates with full auditability.
- Unified data and model serving inside Databricks lakehouse
- Reduced model deployment latency, enabling millisecond real-time pricing
- Governance and traceability enforced through Unity Catalog
Infrastructure signal
ERGO Hestia replaced a fragmented pipeline that exported pricing data from Databricks to an external Azure PostgreSQL database and then used an adapter with caching for serving. This previous architecture introduced latency and complexity as data volume and model iteration frequency grew, limiting scalability and increasing governance risk in a regulated domain.
The new infrastructure leverages Databricks Lakebase’s transactional relational layer on Delta tables for continuous, automated synchronization of processed data, removing the need for external extraction jobs or databases. Model Serving Endpoints provide direct API access for pricing engines, collapsing multiple steps into a single managed layer inside the lakehouse, simplifying overall deployment surface area and reducing operational overhead.
Developer impact
Developers and data scientists benefit from accelerated workflows by hosting data and model serving within the same ecosystem. Models are logged via MLflow and registered in Unity Catalog, enabling simultaneous testing of multiple model versions against live data with full version control and audit trail capabilities.
This integrated approach removes barriers between training, governance, and deployment, allowing pricing experts to push updates faster and with greater confidence. The native environment supports real-time responsiveness at millisecond scale, which improves development agility and ultimately shortens time-to-market for new pricing innovations.
What teams should watch
Teams focused on cloud cost optimization and reliability should note the elimination of redundant infrastructure components such as external databases and caching layers. Consolidating query execution and data serving into Lakebase reduces cost drivers related to data egress, synchronization jobs, and operational complexity.
Product and platform teams must monitor the impact on observability and governance since Unity Catalog now governs both data and model artifacts. Ensuring seamless traceability of pricing decisions and compliance with regulatory standards will require coordinated workflows between data engineering, ML ops, and application teams using the unified lakehouse platform.