Skip to Content
View all events

Copilot-Ready Power BI: Preparing Semantic Models for AI

Published by O'Reilly Media, Inc.

Intermediate content levelIntermediate

Make your Power BI models readable to Copilot and data agents, and use Copilot properly in day-to-day development

What you’ll learn and how you can apply it

  • Audit an existing Power BI semantic model against what Copilot and data agents actually read
  • Apply naming, description, synonym, measure, and relationship practices that make AI answers accurate
  • Use TMDL and AI-assisted scripting to prepare a model at scale
  • Use Copilot in Power BI Desktop and the service for day-to-day development, and recognize where it does not pay off

Course description

Copilot and data agents do not read your reports. They read your semantic model—its table and column names, its descriptions, its synonyms, its relationships, and the measures you did or did not define. When AI gives a confidently wrong answer about your data, the model is usually the reason. Most Power BI models in production today were built for humans reading dashboards, not for machines answering questions.

In this two-hour session, Microsoft Data Platform MVP and Microsoft Fabric book author Nikola Ilic shows you exactly what AI sees when it looks at a Power BI model, what it misreads, and how to fix it. You’ll work through the preparation that makes Copilot and data agents reliable: description and synonym discipline, measure coverage, unambiguous relationships, and field visibility. You will also see how to do this at scale with TMDL and AI-assisted scripting, rather than one field at a time.

The second half turns to Copilot itself. You’ll see honest, unedited workflows for measure creation, report building, and narrative generation in Power BI Desktop and the service, including the cases where Copilot still costs more time than it saves, and where an external AI tool is the better choice.

Note: The sample model used in the demonstration will be shared as a downloadable resource along with the AI-readiness checklist and the TMDL scripts.

This live event is for you because...

  • You own or maintain Power BI semantic models and have been asked to make them work with Copilot or AI agents.
  • You have tried Copilot or a data agent, gotten unreliable answers, and want to know what to fix.
  • You’re an analytics engineer or BI developer who wants a concrete standard for AI-ready modeling.
  • You want a candid assessment of Copilot in Power BI before committing your team to it.

Prerequisites

  • Working knowledge of Power BI Desktop (building reports, creating measures, and working with Power Query)
  • Familiarity with semantic models, relationships, and star schema concepts
  • No prior experience with AI, Copilot, or agents required

Recommended follow-up:

Schedule

The time frames are only estimates and may vary according to how the class is progressing.

Your model is now an interface (10 minutes)

  • Presentation: Why the semantic model became an AI consumption surface, and what that changes about how you build one
  • Q&A

What AI actually sees in your model (25 minutes)

  • Presentation: How Copilot and data agents read a semantic model—metadata, descriptions, synonyms, relationships, and measure definitions; what’s invisible to AI and what actively misleads it—cryptic column names, unhidden key columns, ambiguous relationships, missing date tables; failure gallery—the same question asked against an unprepared and a prepared model
  • Q&A

Preparing the model, at scale (45 minutes)

  • Presentation: The preparation checklist in priority order—what to fix first when you cannot fix everything; naming and description discipline, synonyms and the linguistic schema, field visibility, and measure coverage; context engineering—writing descriptions that answer the questions AI will actually ask
  • Demonstration: Taking a raw semantic model to AI-ready end to end; using TMDL and AI-assisted scripting to apply the prep across hundreds of fields instead of one at a time
  • Q&A
  • Break

Copilot in day-to-day Power BI development (30 minutes)

  • Presentation: Copilot inside Power BI honestly assessed—where it adds real value and where it wastes time; practical prompting patterns for Power BI Desktop and the service
  • Demonstration: Real workflows for measure creation, report building, and narrative generation, including the attempts that do not work; how native Copilot differs from external AI chat tools, and when to reach for each
  • Q&A

Making it stick (10 minutes)

  • Presentation: Turning the checklist into a team standard—what to enforce in review and what to automate; where to go next
  • Q&A

Your Instructor

Nikola Ilic

Nikola Ilic is known as “Data Mozart,” because he’s making music from the data! He works as an independent data platform consultant and trainer focusing on Microsoft technologies. For his exceptional contributions to the tech community, Nikola was awarded a Microsoft Data Platform MVP. He’s also a Microsoft Certified Trainer, blogger, public speaker, mentor, and author of many online courses. He’s interested in everything related to data and is always eager to extract valuable info from raw data most effectively.

Skills covered

  • Power BI
  • Artificial Intelligence (AI)
  • Scikit-learn
  • Machine Learning