Almanac
Microsoft/power-biPower Platform

Consultant KB for Microsoft Power BI: semantic models, DAX and modelling, Power Query and dataflows, reports and visuals, the service and workspaces, capacity and performance, embedding and integration, governance and security, ALM and deployment, and licensing, plus cross-cutting decision guides. Scoped to Power BI, with Microsoft Fabric covered where it touches Power BI directly. Implementation notes, configuration decisions and the gotchas that bite on real projects. Populated by the daily author agent from the Power BI release plans, docs repo and product blog, plus the author's own consultant notes.

power-bi/05-service-and-workspaces/feature-copilot-in-power-bi.md
feature-copilot-in-power-bi.mdv1 · history
CurrentApplies to BothUpdated 3 days agoSource Microsoft Learn ↗

What it does

Copilot in Power BI is a family of AI experiences rather than one feature: the Copilot pane on a report, the full-screen standalone Power BI agent, Copilot inside an org app, and Copilot in web modelling. Alongside those, Microsoft 365 Copilot can now answer questions from Power BI reports and semantic models directly in Copilot Chat, grounded through Fabric IQ.

Key facts

  • Status varies by surface. The report Copilot pane is GA. The standalone Power BI agent, the app agent and Copilot in web modelling are in preview.
  • Data answers in Microsoft 365 Copilot Chat went GA in the September 2026 feature summary: users can find a report or model, attach it, and ask questions in natural language. Microsoft 365 Copilot inside Power BI itself is a separate preview announced in the same release.
  • Copilot needs organisational capacity: a paid Fabric capacity of F2 or above, or Power BI Premium P1 or above. A Pro or PPU licence on its own isn't enough, and trial capacities and free SKUs don't qualify.
  • After buying or scaling capacity it can take up to 24 hours for Copilot to appear.
  • The tenant setting Users can use Copilot and other features powered by Azure OpenAI must be on. The standalone and app experiences need it at tenant level; capacity-level delegation isn't enough. The standalone agent also needs its own preview setting.
  • Tenants or capacities outside the US or EU data boundary have Copilot off by default until an admin allows data to be processed outside the geography.
  • Sovereign clouds aren't supported, and the standalone and app agents aren't available in a list of regions that includes India West, UAE Central, Germany North and Norway West.
  • If a Fabric Copilot capacity exists, all Copilot usage bills to it. Without one, standalone Copilot auto-selects a Copilot-enabled workspace to bill against.
  • Usage is token-driven and runs as background capacity operations. The Fabric Capacity Metrics app shows it under the Copilot in Fabric operation.
  • An identical prompt against an unchanged model within a rolling 24 hours returns the cached answer. Clear chat doesn't reset that.
  • Since September 2026 a semantic model setting lets editors block read-only users from reaching the model through Copilot, Microsoft 365 Copilot Chat and Cowork, data agents, and the remote Fabric and Power BI MCP servers. It's on (access allowed) by default.

When to use / skip

If the client already has F2+ or P1+ capacity, switch the report pane on for a pilot group and see what people actually ask. It's the cheapest way to find out whether their models are fit for natural-language questions. Skip the standalone agent for production use until it leaves preview, and don't sell Copilot to a Pro-only estate: the capacity requirement turns a licensing conversation into a platform one.

Configuration decisions

  • Which capacity Copilot bills to: a dedicated Fabric Copilot capacity, or whatever workspace the service auto-selects. The default makes cost hard to attribute.
  • Who gets Copilot, through tenant-setting security groups rather than everyone at once.
  • Whether cross-geo processing is acceptable for non-US/EU tenants. That's a data protection decision, not an IT one.
  • Per semantic model, whether read-only users should reach it through Copilot and MCP at all.
  • Whether org apps show Copilot in their navigation (per-app setting, on by default).

Gotchas

  • The model is the product. Copilot answers from the measures, relationships and descriptions you've defined, so a model with cryptic names and no descriptions gets poor answers and the user blames Copilot.
  • The 24-hour cache catches testers out. Re-running the same prompt after a fix "still shows the old answer" until the model refreshes or the wording changes.
  • Auto-selected billing workspaces can land Copilot load on a capacity that's already busy with refreshes, and throttling then hits reports that have nothing to do with Copilot.
  • Non-English prompts sometimes work but aren't officially supported. Don't promise multilingual Copilot to a European rollout.
  • The read-only block covers Copilot and MCP access only. It isn't a substitute for row-level security or workspace permissions.

Consultant notes

  • Run a model readiness pass before any Copilot demo: business-friendly names, descriptions on key measures, hidden helper columns. It's the difference between a good demo and an awkward one.
  • Put the Fabric Copilot capacity decision in the design document. Finance will ask where the CU spend went, and "it picked a workspace" isn't an answer.
  • For regulated clients, walk through the cross-geo setting and the new read-only control with the data protection officer early. Both are easy to flip and hard to explain afterwards.
  • Expect Microsoft 365 Copilot to become the main entry point for casual users. Design models that answer well there, not only inside Power BI.

Worth revisiting when the standalone agent or Microsoft 365 Copilot in Power BI reaches GA

Was this accurate?
Related
Fabric capacity and F SKUs06-capacity-and-performanceFabric capacity licensing for Power BI10-licensingPower BI KB — Features & Changelog_rootAgentic development with skills for Fabric09-alm-and-deploymentPower BI licensing overview10-licensingXMLA endpoint09-alm-and-deployment