A shared set of business definitions between raw data and reports, so every tool computes a measure the same way.
A semantic layer sits between the database and the dashboards, spreadsheets and AI assistants that read it. It maps raw tables and columns to business terms, such as net revenue, active customer and region, and defines each measure once. Every tool that queries through it gets the same answer to the same question.
A semantic layer is a set of definitions, not a formula. Each measure is written once: its inputs, filters, joins and aggregation, for example net revenue = gross invoiced − credits − intercompany sales.
Gross invoiced revenue of $3,500,000, less $180,000 of credits and $120,000 of intercompany sales, gives net revenue of $3,200,000. With the definition in the semantic layer, the dashboard and the finance spreadsheet both show $3,200,000 instead of one showing $3,500,000.
Definitions that change without a version, so last quarter's figure no longer matches, and teams that bypass the layer with their own extracts. For how BI tools handle it, see business intelligence tools.