Overview
A regional property and casualty insurer was replacing its core systems with a ground-up Guidewire implementation spanning PolicyCenter, BillingCenter, and ClaimCenter. The analytics work began with a single directive: build one model that serves the whole enterprise. Guidewire’s schema is large, effective-dated, and polymorphic, with many record types sharing one table. It also differed between environments. The modeling approach was still undecided, with Data Vault among the options. Meanwhile, analysts, executives, and downstream teams depended on existing reports that had to keep working through cutover. The insurer brought in evolv to design the enterprise analytics platform and deliver it on the implementation’s timeline.
action plan
Over 8 months, an evolv data architect partnered with the insurer’s data team to design a leaner alternative to Data Vault (a full SCD-2 historical layer beneath a Kimball star schema) and wrote the modeling standard that let Cortex agents handle the repetitive authoring while the architect owned the design.
Steps
1
Built the foundation from metadata: Python generators turned every staged Guidewire table into about 3,400 dbt source definitions and SCD-2 historical models, all run from one codebase across development, UAT, and production.
2
Encoded the standard in a 10-skill Cortex library and converted Guidewire’s data dictionary into a Snowflake source of truth the agents could query, with one firm rule: no build without traceable lineage to source.
3
Gated every model with about 17,000 generated tests, grain validation, and source-to-model reconciliation, while an observer loop flagged issues in both the architect’s and the agents’ work and fed each fix back into the standard.
Results
4–7 Months of Modeling Work Saved
Across about 120 dimensions and 90 facts, Cortex agents cut a complex fact from 2–3 days by hand to 1–2 hours including review, saving roughly 650–1,200 engineering hours.
Zero Report Disruption at Cutover
50 analytics views blend new Guidewire data with legacy data, so 300+ reports and their dashboards kept running through cutover.
Built to Extend
Phase 1 modeled the core business processes with every field from each source table, so thousands of additional columns are one view-layer change away from any report. The insurer’s data engineer now owns the standard and skill library, and new facts or dimensions take hours, not days.
Conclusion
“evolv’s dimensional model is a dream to work with. I built our analytics views and a semantic layer for AI directly on top of it, and it’s been easy to extend.” — Data Engineer, regional P&C insurer