Power BI measures are your company’s brain. Here’s how AI finally uses them.

Copilot showed the demand — and the compromise

Microsoft’s own Copilot proved enterprises desperately want to ask their data questions in plain language. It also proved the compromise they are asked to make: the answers happen in Microsoft’s cloud, under Microsoft’s terms, with trust-it agreements about training data. A large payments company we know responded exactly the way regulated companies respond: they blocked every AI tool except the one with a no-training promise — and their staff still dislikes using it. The demand is real. The trust model is the problem.

The enterprise answer is not a better promise. It is an architecture where no promise is needed: the model sees shapes and definitions, never rows; the answers assemble inside your environment; the audit line says what happened. When the boundary is enforced, “will they train on our data?” stops being a question.

Your semantic model is AI context, already written

A Power BI semantic model is everything an accurate AI needs and nothing it should fear: measures with exact definitions (the “80 not 100” rules, written in DAX), dimensions with business names, relationships between tables, and security roles that already encode who may see what. It is a certified definition of your business, maintained by people you pay to keep it right.

The technical path to read it exists and is documented: TMDL (the Tabular Model Definition Language) and the XMLA endpoint expose the model’s metadata to machines. An AI layer that ingests those definitions inherits your analysts’ decade of work in an afternoon — and stops making the mistakes that come from re-deriving the business from raw tables.

The pattern for Power BI shops

Inherit the model: measures and dimensions become certified definitions, reviewed once by the people who own them. Keep the roles: the same row-level logic your reports enforce applies to every AI answer — a branch manager sees their branch, never the whole company. Answer inside: values never leave; only shapes and definitions travel. Prove it: one audit line per answer, so the compliance conversation is a printout, not a project.

And the queue dies. The 10–14 day wait for a new report — the one that exists even in the most data-mature Power BI shops — collapses into a question asked in Arabic or English and answered in seconds, on definitions your own team certified.

DEBO connector status: P1, on the roadmap

Honest badge: the Power BI connector (TMDL/XMLA semantics → certified definitions) is in the first connector batch. Enterprise Power BI shops are exactly who it is being shaped for — if that is you, early conversations shape the connector.