An enterprise data warehouse (EDW) is a single, organization-wide store of integrated, historical data covering every subject area - sales, finance, supply chain, customers - under one set of definitions. A data mart is a smaller slice scoped to one subject area or one department, holding only the data that audience needs. The two aren't rivals so much as different altitudes: the EDW is the company's whole picture, a mart is the marketing team's or the finance team's view of it. Where they sit relative to each other is the design decision.
Dependent vs independent marts
A dependent data mart is sourced from the EDW. The warehouse integrates and conforms the data once, and the mart is just a subject-area projection of it - so "revenue" means the same thing in the finance mart and the sales mart, because both descend from the same conformed source. An independent data mart pulls straight from operational systems on its own, bypassing any central warehouse. Independent marts are faster to stand up and are how many "we'll just build one for marketing" projects start, but every mart redefines its own keys and metrics. Five independent marts give you five definitions of "active customer" and no way to reconcile them - the classic data-silo failure.
Two ways to build the warehouse
The order you build these in is the long-running Inmon-versus-Kimball debate. Bill Inmon's EDW-first approach builds the normalized enterprise warehouse as the foundation, then spins dependent marts off it - top-down, integration before delivery. Ralph Kimball's bottom-up approach builds dimensional marts first, one business process at a time, tied together by conformed dimensions so they compose into a virtual enterprise view without a separate normalized layer.
| Aspect | EDW-first (Inmon) | Marts-first (Kimball) |
|---|---|---|
| Build order | Central EDW, then marts | Marts, unified by conformed dimensions |
| Central model | Normalized (3NF) | Dimensional bus |
| First value delivered | Later | Sooner |
| Integration risk | Lower (designed in) | Depends on conformance discipline |
See the Inmon method and Kimball method for the full treatment. The connection between the EDW and its dependent marts is usually a hub-and-spoke architecture.
Frequently asked questions
What's the difference between an EDW and a data mart? An EDW spans the whole organization with integrated data across all subject areas; a data mart is a subject-area or departmental subset scoped to one audience's needs.
Dependent vs independent data mart? A dependent mart sources from the central EDW, so definitions stay consistent; an independent mart pulls directly from operational systems, which is faster to build but breeds conflicting metrics across marts.
Do I need an EDW if I have data marts? Not strictly - Kimball's bus architecture delivers an enterprise view through conformed dimensions across marts without a separate normalized EDW. The EDW-first path adds one deliberately for integration.
Why are independent marts risky? Each one defines its own keys and metrics, so the same business term ends up meaning different things in different marts, and reconciling them after the fact is expensive.