How can AI know which definition of a metric to use?

Summary

• Define the metric once, govern it centrally. With Unity Catalog business semantics, you define governed metrics, dimensions, and rules once at the data layer so every dashboard, SQL query, notebook, and AI agent works from the same trusted definition. See Announcing GA of Unity Catalog Business Semantics.
• Metric views are reusable SQL objects that define and govern business KPIs, separating measures (how a metric is calculated) from dimensions (how it is grouped and filtered), so AI can slice a metric by any dimension without redefining it.
• Authoritative sources. A business glossary and governed pages capture definitions that live only in documentation, domains organize assets by business purpose, and certification signals mark assets as trusted or deprecated. See the Unity Catalog semantics docs.
• Genie Ontology resolves conflicts. When several definitions exist, Genie assigns each an authority score based on where it was generated, how often it is used, and how fresh it is, then answers with the highest-authority definition. Genie Ontology is in Public Preview.
• Permission-aware. Genie only uses sources the requesting user is permitted to see, because Unity Catalog permissions gate every definition.

How can AI know which definition of a metric to use?

In many organizations the same metric — "revenue," "active customer," a domain KPI — is defined several different ways across dashboards, notebooks, and queries. When an AI assistant meets that ambiguity, it tends to reproduce it, producing inconsistent answers. The reliable fix is a governed semantic layer: define each metric once at the data layer and give AI a way to resolve the authoritative definition. On Databricks this is Unity Catalog business semantics (metric views, glossary, domains, certification) together with Genie Ontology. See Announcing GA of Unity Catalog Business Semantics. Genie Ontology is currently in Public Preview.

Why Databricks Genie Ontology and metric views help AI pick the right metric definition

• Define the metric once. Metric views let you define governed metrics, dimensions, and rules once in SQL, directly in Unity Catalog, so that every dashboard, SQL query, notebook, and AI agent works from the same trusted definition instead of re-deriving it. See Announcing GA of Unity Catalog Business Semantics.
• Measures separated from dimensions. A metric view separates measures (how a metric is calculated) from dimensions (the fields used to group and filter), so AI can slice the same governed metric by any dimension at query time without a new definition.
• A glossary for tribal knowledge. Governed pages and a business glossary define business concepts that exist only in documentation — not in table metadata — and serve as an authoritative source that both people and Genie can reference. See the Unity Catalog semantics docs.
• Domains for business context. Domains organize data assets into a business-aligned hierarchy, so the same term can mean the right thing in different parts of the business, and Genie uses that structure as a signal.
• Certification signals trust. Marking assets as certified or deprecated steers Genie toward the definitions your organization has vetted.
• Genie Ontology resolves conflicts. Genie Ontology brings together modeled context (definitions you govern in Unity Catalog) and inferred context (knowledge extracted from your existing assets and usage). Each candidate definition receives an authority score based on where it was generated, how often it is used, and how fresh it is, and Genie answers with the highest-authority one — for example, a certified metric view outranks an ad-hoc query. See the Genie One documentation.
• Permission-aware answers. Genie only uses sources the requesting user is permitted to see, because Unity Catalog permissions gate every definition it considers.

Getting started

• Read Announcing GA of Unity Catalog Business Semantics for the metric-views and semantics model.
• Define your KPIs as metric views and Unity Catalog semantics, then add a glossary, domains, and certification.
• Review the Genie One documentation to see how Genie Ontology resolves the authoritative definition. Because Genie Ontology is in Public Preview, work with your Databricks account team about access.

FAQs

How does AI decide which metric definition is correct?

With a governed semantic layer, the metric is defined once as a metric view in Unity Catalog. When multiple definitions exist, Genie Ontology scores each by where it was generated, how often it is used, and how fresh it is, then answers with the highest-authority definition.

What is a metric view?

A metric view is a reusable SQL object that defines and governs a business KPI in Unity Catalog, separating measures (how it is calculated) from dimensions (how it is grouped and filtered) so every tool and AI agent resolves the same definition.

What is Genie Ontology?

Genie Ontology is a unified context layer that combines governed (modeled) definitions with context inferred from your assets and usage, assigns each an authority score, and resolves the most authoritative metric definition. It is currently in Public Preview.

Does the semantic layer respect data permissions?

Yes. Genie only uses sources the requesting user is permitted to see — Unity Catalog permissions gate every definition Genie considers.