Why centralize metric definitions?

“Active customers” can produce different numbers across dashboards if teams use different activity events, time windows, or deduplication rules. A metrics layer moves the definition out of individual reports and makes it reusable by queries, applications, and agents.

What does the layer provide?

A metric definition can include its business description, formula, source measure, filters, time grain, available dimensions, owner, and version. The layer may generate a query or expose the result through an interface. For ratios such as conversion rate, the numerator and denominator must use compatible scopes, with a stated rule for zero denominators.

For example, monthly active customers may mean unique customers with at least one valid transaction during a calendar month. A different concept, such as active leads, needs its own name and definition. Shared definitions make comparisons meaningful across teams.

Where does it end and the semantic layer begin?

The metrics layer focuses on metric definitions and relevant dimensions. A semantic layer also models entities, relationships, grain, and access rules. Product boundaries vary. Agents additionally need business scope, valid breakdowns, and permissions. Defining business semantics in the data engine gives an example.

FAQ

Does it store the metric’s data?

It usually stores definitions; data stays in source systems or warehouses.

Can metrics have multiple valid versions?

Yes, when scopes or policies differ, but each needs an explicit name and owner.

Who approves changes?

A designated business owner with data engineering review.