Most analytics problems aren’t tool problems — they’re modelling problems: nobody knows exactly how a metric is built, and every team calculates it differently. Analytics engineering tackles this head-on, treating data as a versioned, tested product.
The layers
A proven pattern separates three levels: Raw (source data, untouched), Staging (cleaned, typed — one source, one model) and Marts (business-facing models for reports). BigQuery is the warehouse, dbt the discipline on top.
Why dbt makes the difference
- Transformations as code — SQL in Git, with review and history.
- Tests —
not_null,unique,relationships: assumptions become verifiable, not hoped for. - Documentation & lineage — every column has a traceable origin.
The real win
When a metric is defined in exactly one place, tested and documented, the arguments about “the right number” stop. That’s the goal: not more dashboards, but dependable ground for decisions.