Overview
- Metric definitions belong in a governed semantic layer, not in dashboards.
- BigQuery is the single computation home; Looker is the governed access surface.
- AI analytics should consume the same semantic layer as human analytics.
Semantic Layer
Define metrics once, govern them centrally
LookML models capture metric formulas, grains, allowed joins, and ownership on top of BigQuery datasets, so every dashboard and report computes the same numbers from the same definitions.
- Version metric definitions and review changes like code.
- Deprecate metrics explicitly with owners and migration notes.
Performance
Design BigQuery for predictable cost and latency
Partitioning, clustering, materialized aggregates, and scan budgets keep interactive dashboards fast and monthly bills explainable as usage grows.
- Partition by business date and cluster by the most-filtered dimensions.
- Serve high-frequency dashboards from materialized aggregates, not raw fact scans.
One Surface
Humans and AI share the same governed surface
When Gemini NL2SQL, scheduled reports, and Looker explorations all resolve through the same semantic layer and permissions, answers stay consistent no matter who — or what — asks the question.
- Route AI-generated queries through governed views, not raw tables.
- Reconcile AI answers against Looker numbers in evaluation checks.