What is data modeling (Inmon, Kimball, Data Vault)?
Data modeling is the structured design of how data is organized, related and stored in a data warehouse or lakehouse. The three most common approaches are Inmon (a normalized enterprise core), Kimball (dimensional star schemas for reporting) and Data Vault (a flexible, historized integration approach).
Also known as: data modeling · dimensional modeling · star schema · fact table · dimension table
Where data modeling is used
Data modeling defines which tables and relationships data is stored in so that it is understandable, consistent and fast to query. A good model largely determines how quickly and reliably reports and analyses can later be built.
The three approaches set different priorities. Inmon (top-down) first builds a highly normalized, enterprise-wide core and derives subject-specific data marts from it; this promotes consistency but is more effort to build. Kimball (bottom-up) models dimensionally in star or snowflake schemas with fact tables and dimensions; this is directly reporting- and Power BI-friendly. Data Vault separates hubs (business keys), links (relationships) and satellites (descriptive, historized attributes); this is especially flexible, well suited to historization and robust against source changes.
A practical example
In the dy Project AG data platform, a large construction project worth over 1 billion CHF, data from SQL Server, Excel and REST APIs was integrated on Azure Databricks. A Data Vault-like approach helps integrate many changing sources flexibly and with history, while the gold layer provided for Power BI is modeled dimensionally in the Kimball style so that reports stay fast and understandable.
Fact tables & dimensions
Two table types are central to the dimensional (Kimball) model:
- Fact table: holds the measurable metrics of a business event — such as revenue, quantity or cost — together with references to the related dimensions.
- Dimension (dimension table): provides the descriptive context you analyze by — e.g. time, customer, product or region. Dimensions answer the by what of an analysis.
- Star schema: the layout where one central fact table connects directly to several dimension tables — simple, performant and ideal for Power BI.
How it relates & how smiit uses it
The three approaches are not mutually exclusive but are often combined: Data Vault for flexible, historized integration of many sources, a dimensional Kimball model for the reporting-facing delivery layer, and Inmon as the guiding idea of a consistent enterprise core. They sit one level below the architectural idea, such as the medallion architecture, which defines in which stages data is refined. smiit chooses the approach based on source landscape, rate of change and reporting requirements and combines them pragmatically rather than following one camp dogmatically.
Common mistakes & misconceptions
- Data modeling is not just creating tables; it defines entities, relationships and granularity and largely determines performance and analytical usefulness.
- Many think a single flat table model is simplest, but in analytics a star schema with facts and dimensions is usually clearer and faster.
- A common error is to assume normalization is always the goal. For reporting and BI, deliberate denormalization is often more useful than a strictly normalized model.
Frequently asked questions
Which modeling approach is the best?
There is no universal best. Kimball is reporting-friendly, Data Vault is strong with many changing sources and historization, and Inmon delivers a consistent enterprise core. In practice they are often combined.
What is the difference between Kimball and Data Vault?
Kimball models dimensionally in star schemas for direct reporting. Data Vault separates hubs, links and satellites for flexible, historized integration and is often used as a layer before a dimensional model.
What are a fact table and a dimension in simple terms?
A fact table holds the measurable values of a business event, such as revenue or quantity. A dimension provides the descriptive context you analyze by, such as time, customer or product. In a star schema, one central fact table connects directly to several dimensions.
How do data modeling and the medallion architecture relate?
The medallion architecture defines in which stages (bronze, silver, gold) data is refined, while data modeling defines how the tables within those stages are structured. In practice the gold layer intended for reporting is often modeled dimensionally.
Related terms
Sources & further reading
Want to put this topic to work in your company?
Updated · Back to the glossary