A semantic model is a curated layer between your raw data and the tool querying it — whether that's a BI dashboard or an AI agent. It defines which tables matter, how they relate to each other, and what the metrics mean.

For agentic analytics, the semantic model typically consists of 4 to 8 well-modeled tables per business area. A support team, for example, might use dim_accounts, dim_subscriptions, fact_usage_daily, and fact_support_tickets. Four tables with clear relationships and definitions that the agent can use to write accurate queries.

The key insight: agents perform better when pointed at a small, curated set of tables than when given access to the entire data warehouse. Less is more.