A semantic layer sits between warehouse tables and BI users. It defines business-friendly names, metrics, joins, and certified definitions so people do not each reinvent "revenue" in Looker/Tableau/Power BI.
Warehouse tables (fct / dim / marts)
|
Semantic layer (metrics, dimensions, joins, row access)
|
BI tools / notebooks / embedded analyticsWhat it usually encodes
- Certified metrics:
gross_revenue,net_revenue,active_customers - Join paths from facts to dims
- Synonyms and descriptions
- Sometimes row-level security
Why it matters
Without it, five dashboards define "active user" five ways. The semantic layer is the contract for metric consistency, complementary to dimensional models, not a replacement for grain.
Tools people mention
LookML, dbt MetricFlow / Semantic Layer, Cube, AtScale, Power BI datasets, Tableau published data sources.
Interview tip
> "Semantic layer = governed business definitions on top of tables so metrics stay consistent across tools."