Conformed Dimension

A conformed dimension is a dimension whose attributes mean the same thing in every fact table that uses it. The Kimball Group’s definition is concrete: dimension tables conform when their attributes have “the same column names and domain contents.” Orders, shipments, and returns each have their own fact table; they can be compared by customer or by month only if customer and month are conformed across all three.

What conforming requires

Matching column names is the easy part. Domain contents — the actual values — have to match too. A region labeled “APAC” in one table and “Asia-Pacific” in another does not conform even if both describe the same territory, because a report grouping across both will show two separate rows. The same applies to codes, hierarchies, and definitions: a fiscal month and a calendar month under the same column name do not conform.

When they do conform, measures from separate fact tables can be placed side by side by querying each one separately and aligning the results on the shared attributes, which is what drilling across means.

Conforming at different grains

Not every fact table works at the same level of detail. A sales forecast might be made per month and brand, while actual sales are recorded per day and product. The Kimball Group handles this with shrunken dimensions: conformed dimensions that are a subset of the rows and/or columns of a base dimension. A month dimension whose attributes are exactly the month-level attributes of the date dimension still conforms with it, so forecast and actuals roll up to the same months without a translation step.

Why it is an organizational problem

The Kimball Group describes conformed dimensions as defined once, in collaboration with the business’s data governance representatives, and then reused across fact tables. The technical structure is simple; the agreement is not. Deciding what “customer” means across sales, billing, and support is a negotiation with owners who each have a legitimate definition for their own purpose.

When that agreement is skipped and each process keeps its own copy, the divergence surfaces as apparent data quality problems. In an invented example, one report counts 1,200 active customers and another 1,190; the investigation eventually finds that one copy excludes accounts closed during the month. That reconciliation recurs every reporting cycle until the definition is agreed. Keeping a dimension conformed in practice needs a named owner for the definition, a process for changing it, and one load path that every consumer reads from.

Conforming does not require one meaning for every word across the organization. It requires one meaning for the attributes that are deliberately shared — which is why some terms are conformed and others are left to their own bounded contexts. How conformed dimensions keep cross-process totals honest is worked through in Grain, Facts, and the Arithmetic That Quietly Goes Wrong.

References: Kimball Group, Conformed Dimensions; Kimball Group, Shrunken Rollup Dimensions.