Data Engineering › Data Modeling for Analytics

Dimensional Modeling

Kimball's approach: facts surrounded by descriptive dimensions.

Data Engineer · Mid-level · core Data Analyst · Mid-level

Also known as: Kimball dimensional modeling, Kimball method, dimensional model, dimensional modelling, star schema design

Dimensional modeling, popularized by Ralph Kimball, is the classic way to design analytical databases so they’re easy to understand and fast to query. The data is organized into fact tables (measurable events) surrounded by dimension tables (the descriptive context), usually as a star schema.

            dim_date
               │
 dim_customer ─┼─ fact_sales ─ dim_product       "revenue (fact) by month, region, category (dimensions)"
               │
            dim_store

The way business users think (“sales by product and region by month”) maps directly onto it.

Kimball’s four-step design process

Choose the business process to model: order fulfillment, web sessions, support tickets.
Declare the grain: exactly what one row of the fact table means (“one row per order line”) (grain). Do this before anything else.
Identify the dimensions: the who, what, where, when and how that describe each fact (dimension tables).
Identify the facts: the numeric measurements at that grain (fact tables).

Key ideas

Conformed dimensions: a shared dim_customer or dim_date used by many fact tables, so different processes can be analyzed together consistently (conformed dimensions, bus matrix).
Surrogate keys for dimensions (surrogate keys).
Slowly changing dimensions to keep or overwrite history of attributes (SCD).
Different kinds of fact tables: transaction, periodic snapshot, accumulating snapshot (fact table types).
Denormalized dimensions, favoring simplicity over storage savings.

Why it endures

Business-friendly: analysts and BI tools understand it quickly.
Consistent metrics through shared dimensions and defined grain.
Performs well on analytical engines.
Stable under change: new attributes and new facts can be added without redesigning.

Alternatives and context

Inmon’s approach models a normalized enterprise warehouse first (Inmon vs Kimball), and Data Vault focuses on auditability and flexible integration (Data Vault).
Some modern teams skip heavy modeling for wide tables (one big table) when it suffices.
Dimensional models are typically built in a layer on top of cleaned data (data transformation).

Even if you never build a full Kimball warehouse, the habit of declaring the grain and separating facts from dimensions improves any analytical table.