What is a semantic layer in business intelligence?

Summary

• A semantic layer translates raw source data into shared business meaning. It defines metrics, dimensions, and governed definitions so every downstream surface, from dashboards to notebooks to AI tools, uses the same numbers.
• It creates a single source of truth. When "revenue" or "active customer" is defined once, every query relies on the same approved definition instead of teams getting different answers for the same metric.
• On Databricks, the semantic layer is delivered through Unity Catalog metric views — governed, reusable objects that define business KPIs as first-class assets with measures, dimensions, and joins.
• Define once, use everywhere. A metric defined in Unity Catalog is available across AI/BI dashboards, Genie, notebooks, SQL, and jobs, with consistent results wherever it is queried.
• Governed by default. Metric views inherit Unity Catalog permissions, lineage, audit, and row- and column-level security.

What is a semantic layer in business intelligence?

A semantic layer is the architectural component that translates raw source data into shared business meaning. It defines metrics, dimensions, and governed definitions that enable consistent data access across every downstream surface, including dashboards, query editors, data science notebooks, and AI-powered tools. Without a semantic layer, organizations face fragmented metric definitions: two teams ask for the same metric and get different numbers, and AI tools can contradict finance reports. The semantic layer solves this by creating a single source of truth, so that when a term like "revenue" or "active customer" is defined in one place, every query relies on the same approved definition. See Semantic layer architecture: components, design patterns, and AI integration.

How Databricks AI/BI delivers a semantic layer

Databricks provides its semantic layer through Unity Catalog metric views — governed, reusable objects that define business KPIs as first-class assets in the lakehouse.

• A metric is defined once. A metric view specifies measures (how a metric is calculated, such as SUM(order_amount)), dimensions (fields used to group and filter, such as region or order date), and joins (relationships between tables), along with semantic metadata like display names, formats, and synonyms. See Unity Catalog semantics.
• Definition is separated from consumption. A metric view separates the definition of a metric from its consumption, so users receive the same calculated result whether they query through a SQL editor, a dashboard, or an AI agent.
• Define once, use everywhere. Once a metric is defined in Unity Catalog, it is accessible across AI/BI dashboards, Genie, notebooks, SQL queries, and Lakeflow jobs, which removes logic lock-in where a definition lives only in one dashboard or in people's heads.
• Governed by default. Metric views inherit Unity Catalog permissions, lineage, and audit trails, and row-level and column-level security apply automatically, so business logic is centralized, versioned, and auditable.

How AI/BI dashboards and Genie use the semantic layer

• AI/BI dashboards query metric views through a SQL warehouse so that every visualization uses consistent, governed definitions. Dashboard designers can drag and drop predefined metrics instead of writing SQL for each widget.
• Genie, the natural language query interface, integrates directly with metric views. It uses semantic metadata such as synonyms and lineage to map business terms to the right authoritative data, rather than relying on keyword matching alone.
• Dashboard-local metric views let teams define semantic logic directly within an AI/BI dashboard for team-specific or experimental metrics, and a local metric can be promoted to Unity Catalog when it is ready for broader use. See Local metric views.
• Materialization (Preview) provides automatic pre-aggregation with intelligent query rewriting to accelerate metric queries.

Getting started

1. Identify the shared KPIs your organization needs to define consistently, such as revenue, orders, or active customers.
2. Create a Unity Catalog metric view that specifies the measures, dimensions, and joins, plus display names and synonyms.
3. Build AI/BI dashboards on the metric view and let business users query it in natural language through Genie.
4. Review the guidance in Unity Catalog semantics.

FAQs

Why does a semantic layer matter for AI?

Without a governed semantic layer, AI agents reproduce ambiguity and can return inconsistent answers. With one, agents understand not just what data exists but what it means in your business context, and they use approved definitions.

What is the difference between a metric view and a SQL view?

A metric view separates the definition of a metric from how it is consumed and carries semantic metadata, so the same metric returns the same result across SQL, dashboards, and AI tools. A standard SQL view does not provide that governed, reusable metric definition.

Does Databricks ship a prebuilt business glossary?

No. The semantic layer does not come with prebuilt business definitions; customers define their own metrics in Unity Catalog metric views, which then inherit the platform's governance.