Wednesday, November 18th, 8:30am PT

SqlDBM + Engaging Data

Teams are connecting AI assistants and agents to their data warehouses, and many get answers that sound confident but are wrong. The model usually isn’t the problem. The problem is what the AI can’t see: what “active customer” means, which table is the source of truth for revenue, and which definitions the business has actually agreed on.

Confidently wrong: what we found in testing

Serge gave an AI analyst six everyday business questions against the same Snowflake data, once with only the raw schema and once with a semantic model on top. Without the semantic model, it got 2 of 6 right. It reported net revenue at about a third of the real figure, made up its own definition of a “key account,” and counted customers on credit hold as active. With the semantic model, it got 6 of 6 right, with one query per question.

Simon sees the same pattern in the field. At one client, adding a semantic model behind Copilot cut the analysis of three years of energy-consumption data from about 30 minutes to about one minute.

What you’ll learn:

• Why AI answers break down. Undefined business terms, ambiguous metrics and guesses based on column names, and where those gaps usually are.
• The context layer, end to end. How data models, semantic layers and catalogs fit together, with a live walkthrough from a governed model to a semantic layer to an AI assistant answering through MCP.
• What an MCP server actually does. How it gives assistants governed definitions on demand instead of leaving them to infer from raw schemas.
• What’s happening in the enterprise. Simon on industry trends in AI and semantic models, a real semantic layer build from star schema to Power BI and Copilot, and what needs to happen for AI to scale.
• Making your models AI-ready. Practical steps for naming, descriptions, ownership and change control.

Who should attend

Data architects, analytics engineers, data governance leads and data platform owners who are putting AI on top of their warehouse.