A multi-agent approach to audience intelligence

This post introduces an AI-powered audience segmentation solution built on Databricks that uses a multi-agent architecture to bridge the gap between advertising strategy and data execution.

• •A Supervisor Agent (Agent Bricks) orchestrates two sub-agents: a Genie Space agent for natural language-to-SQL audience queries, and an Affinity Agent that computes statistical patterns using Unity Catalog tools.
• •Genie translates natural language briefs (e.g., "affluent travelers aged 35-54") into governed SQL queries against population tables spanning hundreds of millions of consumers and thousands of attributes.
• •The Affinity Agent surfaces statistically validated lift calculations, revealing non-obvious behavioral patterns such as 2.5x cryptocurrency investment rate or 5.7x fitness commitment for a given segment.
• •Data teams encode institutional knowledge once via curated example queries, column comments, and text instructions in the Genie Space, scaling expertise across the entire organiza