Strata
Strata is the data system of record for a national healthcare insurer covering eighteen million lives. We replaced an opaque chain of warehouses and ETL accretions with a governed semantic layer that makes the meaning of the data visible alongside the data itself.
01 — Problem
The client had three competing definitions of 'active member' and seven of 'claim'. Every executive report carried an asterisk, and every analyst spent more time defending numbers than producing them.
Below this was a stack of 4,000 dbt models nobody could fully describe.
02 — Approach
We treated semantics as architecture. The first deliverable was not a pipeline but a domain model — written in collaboration with actuarial, clinical, and regulatory leadership and ratified before any code was changed.
Pipelines were then rebuilt against this model with lineage, tests, and ownership treated as non-optional.
03 — System
Strata is a typed semantic layer over a curated lakehouse, with column-level lineage, a metric registry, and policy-aware access controls. Every dataset has a named owner and a published contract.
Analysts work against the semantic layer, not the storage layer. The two evolve on different cadences.
04 — Outcome
Reporting reconciliation across the executive suite moved from 'roughly aligned' to 'identical to the cent'. Quarterly close shortened by eleven days. The number of analysts able to produce regulator-grade reports independently increased fourfold.
Measured, not asserted.
Quarterly close
−11 days
Lives covered (governed)
18M
Independent analyst capacity
×4