Data Engineering › Storage, Formats & Lakehouse

Data Mart

A subset of the warehouse focused on one team or subject.

Also known as: data marts, mart, subject area

A data mart is a focused subset of a data warehouse built for one team or one subject — sales, marketing, finance. Instead of querying raw tables and joining them by hand, a team queries a small set of clean, modeled tables that already answer their questions.

The classic mistake is letting every analyst query the whole warehouse directly. They each re-implement the same joins and metric logic, get slightly different numbers, and hammer the biggest tables. A mart gives them a governed starting point: for example, a marketing mart with a campaign_performance table rather than six raw event tables.

A mart usually follows dimensional modeling: a fact table of measures joined to dimension tables, often in a star schema.

Dependent vs independent

• A dependent mart is built from the warehouse’s curated data. It inherits consistent definitions and is the common modern pattern.
• An independent mart is fed straight from source systems. It is faster to stand up but tends to duplicate logic and drift from the warehouse.

Trade-offs

Marts make queries simpler and faster, but each new mart adds another place where definitions live. Too many marts, built by different teams, recreate the original problem — conflicting numbers. Keep shared metrics in one place (see the semantic layer) and organize transformation layers deliberately; see staging, intermediate and mart layers.

When not to use a mart: if a single team’s questions are still exploratory, a well-documented set of staging tables may be enough. Build a mart once the access pattern is stable and shared.