Yahoo’s Graph-Powered AI Agent Makes Media Buying a Snap

Yahoo and Google Cloud have teamed up to build a Seller Agent that turns weeks of manual ad campaign planning into seconds of autonomous execution. The secret sauce? A dual-graph architecture that keeps agents grounded in business reality and every decision auditable.

From Weeks to Seconds: The Agentic Leap

Premium digital ad campaigns used to require endless human handoffs, spreadsheets, and manual analysis. Yahoo’s Seller Agent collapses that timeline by letting AI agents plan and execute campaigns in seconds. But speed alone isn’t enough—when real budgets are on the line, you need trust and transparency.

The Dual-Graph Foundation

Knowledge Graph: Grounding Agents in Business Reality

Powered by Spanner Graph, Yahoo’s knowledge graph models advertising products, placements, audiences, contracts, and governance policies as connected entities. Agents navigate this graph to find eligible inventory, apply pricing rules, and check compliance—all in a single query. No more hallucinated deals.

Context Graph: An Auditable Memory

Every action the agent takes is captured in a context graph built with BigQuery Graph. This creates an immutable, queryable record of every decision, score, and policy applied. Auditors can trace a campaign from brief to execution in one query—no more digging through logs.

How It Works: Act and Audit in Parallel

The system runs two simultaneous pipelines:

• Acting via the knowledge graph: The agent submits a brief, retrieves inventory and policies, evaluates candidates, and executes the campaign—all grounded in deterministic business rules.
• Auditing via the context graph: Every decision is logged, creating a closed loop for learning and instant explainability. If an advertiser asks “why this package?”, the answer is a query away.

A Blueprint for Any Industry

This pattern isn’t just for ads. Any high-stakes workflow—financial trading, supply chain, healthcare—can benefit from a dual-graph architecture that grounds decisions in reality and builds an auditable memory. As models become commoditized, your moat is your proprietary graph of business operations and governed history.

Ready to build your own? Start with Spanner Graph for grounding, BigQuery Graph for auditability, and the BigQuery Agent Analytics Plugin to capture traces. For more on building agentic platforms, check out our Google Cloud review.

Original announcement published on Google Cloud.