Dimension table
A table of descriptive attributes — customers, products, dates — that you join to a fact table to slice and label its measures. One row per entity, identified by a surrogate key.
Also: dimensions · dim table
On this page · 6 sections
- Category
- Data modeling
- Also known as
- dimensions, dim table
- Goes deeper in
- Data Modeling
Dimensions are the “by what” of analytics: revenue by customer segment, sessions by device, orders by month. A dimension table holds the descriptive context — names, categories, dates, geographies — that turns raw event measures into something a stakeholder can read.
Dimensions are wide and short relative to facts: many descriptive columns, comparatively few rows. Each row represents one entity (one customer, one product) and is referenced by the fact table through a foreign key, ideally a surrogate key rather than the source system's natural key.
When a dimension's attributes change over time — a customer moves cities, a product changes category — you decide how to track that history. That decision is the slowly changing dimension pattern.
Dimensions are where your business vocabulary lives. A clean, conformed customer or date dimension reused across every fact is what makes metrics consistent company-wide — the same 'customer segment' means the same thing in finance and marketing.
They're also where most of the analytical value sits: rich, well-modeled dimensions let stakeholders slice data in ways nobody anticipated when the fact was built.
- Joining on a messy natural key instead of a stable surrogate key, which breaks when the source reuses or reformats IDs.
- Letting the same concept (e.g. 'region') exist as different, unconformed dimensions across teams.
- Overwriting attributes that change over time when the business actually needs the historical value (use SCD Type 2).
- What is a conformed dimension?
- A dimension that's shared and consistent across multiple fact tables — e.g. one date or customer dimension used everywhere — so metrics line up across business processes.