BigQuery Graphs with measures for trusted agentic workloads

Google Cloud has announced measures in BigQuery Graph, a preview feature designed to enhance agentic workloads by integrating governed metrics with relationship mapping. This update addresses a key challenge where autonomous agents struggle to derive accurate insights from raw tables, particularly when analyzing interconnected business entities. By unifying metrics and graph-based dependencies, the feature allows agents to trace relational paths—such as supply chain disruptions or regional sales variations—with greater precision, reducing the risk of irrelevant or costly recommendations.

The new capability eliminates the need for separate systems to maintain supplier relationships and SQL metrics, which previously required stitching together disparate stacks at runtime. This process was often slow, expensive, and prone to inconsistent KPI calculations. Measures in BigQuery Graph enables mapping existing tables to a property graph in-place with zero ETL, streamlining the workflow. The feature also resolves issues with standard SQL joins during graph traversals, which historically duplicated rows and led to incorrect aggregations, by natively supporting metric definitions within the Property Graph DDL.

Data modelers can now define measures such as SUM or AVG directly within the Property Graph DDL, using standard SQL via the GRAPH_EXPAND function and the AGG aggregator. This ensures the engine resolves structural graph paths before evaluating metrics, allowing agents to determine when to apply calculations (SQL) versus when to analyze relationships (graph). For public datasets like bigquery-public-data, users must map the logical property graph within their own project using a placeholder variable (YOUR_PROJECT_ID), while referencing read-only public tables as nodes and edges.

To simplify the management and deployment of relationship networks, Google Cloud has integrated native operational tools into BigQuery Studio. These include a visual graph modeler—a no-code, drag-and-drop interface for building, editing, and mapping property graphs without manual DDL scripts. Additionally, Looker (LookML) is now natively integrated with BigQuery Graphs as in-database analytic models, enabling businesses to define logic once at the core and avoid maintaining fragmented logic stacks.