Data mart

A data mart is a curated subset of a data warehouse, scoped to the needs of a single business function or team — marketing, finance, sales — rather than the full breadth of the organization's data. It typically contains a small number of fact tables and dimension tables modeled around that team's specific reporting questions.

Data marts can be built in two directions: "dependent" marts are carved out of an existing, already-modeled central warehouse, giving each team a simpler, faster, access-controlled slice of a shared foundation; "independent" marts are built directly from source systems without a central warehouse underneath them, which is faster to stand up but risks each team's numbers drifting apart, since there is no shared source of truth enforcing consistent definitions.

Data marts matter for performance and usability — a marketing analyst querying a warehouse with thousands of tables benefits from a small, purpose-built mart with only the tables and metrics relevant to their work — and for access control, since a mart can be scoped to exclude data a given team should not see. As semantic layers and self-service BI tools have matured, some organizations rely on a governed semantic layer over one central warehouse instead of physically duplicating data into separate marts, though the two approaches are often combined.