Bill Inmon advocates a top-down approach where an enterprise data warehouse is first modeled in third normal form (3NF) across the entire organization, after which departmental data marts are built. Ralph Kimball advocates a bottom-up dimensional bus architecture, building conformed star schemas incrementally around specific business processes.
Inmon enterprise bus vs Kimball marts
Inmon designs the central warehouse as the single source of integrated corporate data, fully normalized to eliminate redundancy and maintain enterprise-wide entity integrity. Departmental teams then extract data from this normalized hub into smaller, focused data marts. This guarantees enterprise consistency but demands massive upfront planning, extensive data governance, and long project delivery timelines.
Kimball bottom-up dimensional design
Kimball focuses on business processes (like billing, inventory, or orders) and builds dimensional star schemas directly. To avoid disconnected data silos, Kimball relies on the enterprise bus matrix and conformed dimensions (standardized dimensions like dim_customer and dim_date shared across fact tables). This allows agile teams to deliver business value within weeks and expand incrementally.
Data Vault as a third option
Data Vault 2.0 has emerged as a popular alternative that bridges this divide. It splits models into Hubs (business keys), Links (relationships), and Satellites (descriptive context and history). Data Vault provides automated scalability, audit trails, and multi-source integration in the raw warehouse layer, from which downstream Kimball star schemas or wide reporting marts are generated.