Why Google Is Extending the Semantic Layer with Business Relationships for Agents

In the past, a Semantic Layer mainly ensured that different BI tools and analysts calculated the same numbers with consistent definitions. With the rise of data agents, the problem shifts to explaining why a metric changes and how that change propagates through business relationships.

1. From "what happened" to "why it happened"

Google Cloud released the paper "Using BigQuery Graphs with measures for trusted agentic workloads" (August 13) to explain why Measures and Graph are combined in a single governed model. This capability was previewed in the April 22 BigQuery update, where Graph native support for Measures was introduced as a "business map" for enterprise agents.

A retail example illustrates the need: a 12% drop in Seattle jacket sales is easy to detect, but understanding the root cause may require traversing multiple hops—orders to distribution centers to suppliers affected by a regional storm. Without a relationship map, an agent might suggest a price cut, missing the supply‑chain issue.

2. Graph finds paths, Measures guarantee numbers

BigQuery Graph with Measures does not simply merge graph queries with SQL aggregates. Because graph traversal can duplicate rows, ordinary SUM or AVG would over‑count. Instead, Measures (SUM, AVG, COUNT, COUNT DISTINCT, MIN, MAX) are defined on node or edge properties and bound to a unique key, ensuring correct aggregation even after expansion. Queries use GRAPH_EXPAND to flatten the graph and AGG to invoke the defined Measure.

Official documentation shows an example: to find the distribution center serving the most distinct customers, the Graph discovers the path User → Order → Product → Distribution Center, while a user_count Measure on the User node prevents double‑counting the same customer.

3. Extending the Semantic Layer to business structure

Google integrates Graph with Conversational Analytics, allowing agents to query the Graph directly, use defined descriptions, synonyms, and Measures, and receive relationship‑aware answers. Agents can generate GQL for traversal or use GRAPH_EXPAND in SQL. Visual Graph Modeler in BigQuery Studio lets users create nodes, edges, and relationships without writing DDL, supporting a zero‑ETL approach that maps existing tables to a Property Graph.

Integration with Looker maps Graph attributes and Measures to LookML, enabling both Looker‑first and Graph‑first workflows while keeping definitions under version control (Git, CI). This unifies KPI definitions across dashboards and agents.

4. Building the business‑context layer

Google Cloud Knowledge Catalog, announced in April, acts as an enterprise Context Engine for agents. It aggregates Measures, LookML, data products, and metadata from various platforms, enriching them to discover entities, relationships, business vocabularies, and validated query patterns. Unlike traditional catalogs that focus on table schemas, this engine supplies business semantics needed to avoid hallucinations in agents.

The overall evolution moves from "Metrics + Dimensions + Joins" to "Metrics + Entities + Relationships + Business Context". The Semantic Layer is not being replaced by a Graph; rather, its boundary expands to include entities and relationships, forming a governed business model that agents can use for root‑cause analysis, cross‑domain decisions, and long‑chain execution.