
BigQuery Measures GA: Unifying BI Metrics and Knowledge Graphs

BigQuery measures is now generally available. Built on BigQuery Graph, it unifies governed metrics with the relational context that explains them, so BI dashboards and AI agents can answer both what happened and why. The post argues that when metric definitions live in business intelligence (BI) platforms and relationships live in knowledge graphs, an agent only sees half the picture and guesses the rest, producing confident but sometimes wrong answers; this is the hallucination problem that BigQuery measures is meant to solve. Defining measures directly on graph entities gives SQL, GQL, dashboards, and agents one shared model instead of siloed stacks, brittle pipelines, and manual reconciliation.
The article illustrates this with a retail bank investigating a spike in transaction activity in the Prague district. Underlying data can stay in native tables or a borderless Lakehouse; a single schema is defined over existing transaction, account, and district tables, with no data movement. Using GRAPH_EXPAND and AGG(), teams flatten the graph and roll up measures, while grain-locking preserves each entity’s native grain. A standard SQL COUNT(account_id) would overcount accounts when joining against thousands of transactions; the measure semantics count each account once. A GQL traversal over the same model identifies which accounts drive the activity, so both halves of the answer come from one governed definition.
The GA release adds richer modeling and semantic discovery. MEASURE() supports composite expressions and derived measures, so baseline aggregates can be composed into higher-order calculations and validated at schema compilation time, catching circular references or invalid aggregations early. A new start_node argument in GRAPH_EXPAND isolates each fact table into its own expansion tree, preventing the chasm trap inflation that happens when multiple fact tables share a common dimension. Data Studio’s BigQuery connector can now visualize graph-native measures directly and honors the model’s aggregation logic, blocking accidental client-side re-aggregations.
For the agentic side, measures can embed business meaning via OPTIONS, including plain-language descriptions, domain synonyms, and custom key-value extensions. BigQuery integrates with Knowledge Catalog so graph and measure metadata is automatically searchable alongside tables, views, and routines, letting stewards, analysts, and agents inspect governance tags, SQL definitions, and business glossaries. BQ.SHOW_GRAPH_EXPAND_SCHEMA returns the expanded graph schema as JSON without scanning underlying tables, flags metrics with is_measure true, and can preserve the entity hierarchy as nested STRUCT rows. The stack integrations include Data Agent Kit for authoring and validating measure definitions from natural-language prompts, LookML-based derived analytic models that compile into BigQuery Graph models, and Conversational Analytics for querying unified graph and measure definitions in plain language.
Woolworths NZ’s Head of Data Enablement, Amanda Reelick, says BigQuery measures lets them deliver self-documenting data products, with Looker Dashboards, Conversational Agents, and their data team queries operating on identical logic. The article closes by noting you can start from an existing BI semantic layer or a knowledge graph and attach measures to the entities you already traverse, building a single source of truth; a video, documentation, and step-by-step codelabs are linked.

