Home / Case studies / Data platform
The data model under a 1.5-billion-record moat
$8B+ alternative asset manager
Valuation runtime
Infrastructure cost, against rising data volume
Claims records in the resulting proprietary data asset
- 013 source systems3 versions of truth
- 02Entity resolutionone record spine
- 03Governed modelquality at ingest
- 04Same-morning answerswas overnight batch
The foundation the underwriting engine and monitoring workflow were built on
The situation
The firm held genuinely rare data — ultimately 1.5 billion claims records — on a platform that couldn’t answer questions at the speed decisions were made. Valuation fit only in an overnight window, so every intraday question became a tomorrow question. And the same entity appeared under different identities in different systems, so any portfolio number required reconciliation before it could be trusted.
What we built
- 01A re-architected data model with entity resolution across the record spine — one asset means one asset everywhere
- 02Platform modernization sized to query patterns rather than storage growth
- 03Valuation pipelines re-engineered from overnight batch to same-morning execution
- 04Data quality instrumentation surfacing defects at ingestion instead of at the point of use
- 05The governed foundation every downstream AI system was subsequently built on
This is the engagement nobody asks for and everyone needs. The two systems that produced the headline numbers were only possible because this existed first — which is why a diagnostic almost always starts at the data layer, not the use-case list.