How Google’s Semantic Layer Gains a Relationship Network: Agents Shift from Metric Calculation to Business Understanding

Historically, the semantic layer’s main task was to ensure that different BI tools, analysts, and applications "calculate the same number" by standardizing metric definitions and dimensions. In the emerging Agent era, the goal evolves: agents must not only know the values of sales, profit, or order counts, but also understand which business objects influence those values and how changes propagate across relationships.

Google Cloud’s August 13 article "Using BigQuery Graphs with measures for trusted agentic workloads" details why governed Measures and Graph are combined in a single analytical model. After an earlier April 22 announcement that introduced native Graph support for Measures in BigQuery, the August piece breaks down three analysis layers:

Metadata Grounding : Identify the available data sources.

Business Metrics : Use Measures to compute performance indicators.

Relationship Mapping : Leverage the Graph to trace why a metric changed.

In practice, Graph + Measure solves a key problem: graph traversal can duplicate rows, leading to over‑counting if a simple SUM or AVG is applied after a join. BigQuery defines MEASURE objects (SUM, AVG, COUNT, COUNT DISTINCT, MIN, MAX) on node or edge properties, binding aggregation to a unique key so that repeated rows are aggregated correctly. Queries expand the graph with GRAPH_EXPAND and then invoke the pre‑defined Measure via AGG. Google likens this to an agent knowing when to use a "calculator" (SQL/Measure) versus a "map" (Graph).

An official example shows how to count distinct customers served by each distribution center without double‑counting after graph expansion, illustrating the practical benefit of Measures in a graph context.

The integration extends beyond raw queries. BigQuery Graph can be consumed directly by Conversational Analytics, allowing agents to ask natural‑language questions that traverse the business map. Google also connects Graph with Looker: Looker can reference an existing BigQuery Graph, mapping its properties and Measures to LookML, or generate Graph DDL from LookML, keeping the model under version control.

Google’s Knowledge Catalog, launched in April, acts as an enterprise Context Engine that aggregates Measures, LookML, data products, and metadata from various platforms, enriching them to discover entities, relationships, and validated query patterns. This contrasts with traditional data catalogs that focus only on table schemas, addressing the "hallucination" risk when agents lack business semantics.

Overall, the shift expands the semantic layer’s boundary from "Metrics + Dimensions + Joins" to "Metrics + Entities + Relationships + Business Context." While the Semantic Layer, Knowledge Graph, and Ontology remain distinct products, the emerging infrastructure provides agents with a governed business model that includes both calculations and the underlying relational context, enabling root‑cause analysis and cross‑domain decision making.