Copilot in Power BI needs a paid Fabric capacity (F2 or larger) or a Power BI Premium capacity (P1 or larger). A Pro or Premium Per User license on its own isn't enough, and trial capacities don't count. For a small business already on Pro, the cheapest way in is one F2 capacity set up as a Fabric Copilot capacity. That lets your Pro users run Copilot on their existing reports and in Power BI Desktop, and the F2 lists at $262.80 a month pay-as-you-go or about $156 a month with a one-year reservation. Whether it's worth that depends far more on your data model than on the license.

What Copilot in Power BI does

Microsoft splits it by who's using it.

People reading reports can ask questions about the data in plain English, get a summary of a report page, and search for reports and models they have access to. There's also a standalone Copilot that works across reports instead of inside one, which is still in preview.

People building reports can ask Copilot to draft a report page, write a narrative visual, write DAX queries, and add descriptions to measures.

People building models can ask it to spot design problems and suggest fixes to measures.

The report-building side is where Copilot earns its keep. Drafting a first version of a page or a tricky DAX query saves real time, and a person checks the result before anyone else sees it. Answering questions for business users is where it can go wrong, because nobody checks the answer.

What you need to turn it on

You need all of these:

  • A paid capacity. F2 or larger, or P1 or larger. Trial capacities and free SKUs aren't supported.
  • The tenant setting. A Fabric admin has to turn on "Users can use Copilot and other features powered by Azure OpenAI" in the admin portal. On paid Fabric capacities it's on by default, but check.
  • A supported region. Copilot runs on Azure OpenAI in US and EU data centers. If your tenant or capacity sits outside the US or EU data boundary, Copilot stays off unless an admin allows your data to be processed outside your region.
  • Patience. After you buy or resize a capacity, Microsoft says it can take up to 24 hours for Copilot to show up.

The standalone, cross-report Copilot needs one more tenant setting turned on, because it's still in preview.

The cheapest setup for a business on Pro

Most small businesses run Power BI on Pro licenses with reports in ordinary Pro workspaces. You don't have to move those reports onto a capacity to get Copilot.

Microsoft has a feature called a Fabric Copilot capacity. You buy a capacity (F2 is the smallest that qualifies), designate it as a Copilot capacity, and assign a group of users to it. Their Copilot usage is billed to that capacity no matter where their content lives. Per Microsoft's documentation, those users can use Copilot in Power BI Desktop and in Pro, Premium Per User and capacity workspaces.

The setup takes three steps:

  1. A Fabric admin enables Copilot for the organization and allows capacity admins to designate Copilot capacities.
  2. The capacity admin designates the F2 as a Fabric Copilot capacity.
  3. The capacity admin assigns a security group of users to it.

Users don't have to do anything after that. Everyone still needs their Pro license to build and share reports. The capacity only covers the Copilot usage.

What it costs to run

Copilot doesn't have its own price. It uses up your capacity's compute, measured in capacity units (CUs).

Microsoft's published rates are 100 CU seconds per 1,000 input tokens and 400 CU seconds per 1,000 output tokens. Their own example request, 2,000 tokens in and 500 out, comes to 400 CU seconds. Copilot runs as a background job, so that usage is spread over 24 hours instead of hitting all at once.

An F2 gives you 2 CUs around the clock, which is 48 CU hours a day. At Microsoft's example size, that's room for a little over 400 requests a day if the capacity is doing nothing else. Real requests vary in size, so treat that as a rough ceiling, not a promise. For a team of 5 to 15 people asking Copilot questions during the day, an F2 is usually a reasonable starting point. Watch the Fabric Capacity Metrics app for the first few weeks and resize if you're running hot.

If you run other work on the same capacity (a warehouse, dataflows, refreshes), Copilot competes with it. Push a capacity past its limit and Fabric starts throttling, which slows everything down. Keeping Copilot on its own small capacity avoids that.

For the full price list and how reservations work, see our post on Microsoft Fabric pricing.

Copilot is only as good as your model

This is the part the marketing skips. Copilot reads your semantic model: table names, column names, measures, relationships. If your model has columns called "Amt2" and "Cust_ID_new", three different revenue measures and no descriptions, Copilot will guess. Sometimes it guesses wrong and presents the wrong number with total confidence.

Before you switch it on for business users, clean up the model:

  • Rename tables and columns to what people call them. "Invoice Amount", not "INV_AMT".
  • Hide what nobody should use. Key columns, helper columns, staging tables.
  • Keep one measure per idea. If there's "Revenue", "Revenue 2" and "Revenue (new)", delete two of them.
  • Add descriptions to measures, especially the ones with business rules behind them.
  • Use a star schema. Copilot handles a clean fact-and-dimension model far better than a pile of flattened tables. Our post on combining tables with relationships covers the basics.

Power BI also has a Prep data for AI button on the Home ribbon in Desktop and on the semantic model in the service. It gives you three tools:

  • AI data schema: pick which tables and columns Copilot should consider, so it ignores the rest.
  • Verified answers: tie common questions ("sales by region this quarter") to a visual you've already checked.
  • AI instructions: plain-language notes for Copilot, like "fiscal year starts in July" or "always exclude internal customers."

Once a model is ready, mark it Approved for Copilot in its settings in the service. That removes the extra warnings the standalone Copilot shows for unapproved models.

Is it worth it for a small business?

Our honest take:

  • Worth it for report builders. If one or two people build most of your reports, Copilot speeds up first drafts and DAX. That alone can justify an F2.
  • Worth it for business users once the model is clean. On a well-named model with verified answers set up for the common questions, people get decent answers without waiting on the report builder.
  • Not worth it as a fix for a messy model. If people don't trust the reports today, Copilot will make that worse, not better.

On data safety, Microsoft's documentation says your data isn't used to train models, prompts aren't stored for abuse monitoring, and Copilot can only see data the person asking already has permission to see. That last point means row-level security still applies. Our post on whether Copilot is safe for client data covers the questions to ask for regulated work.

When to call someone

Two places people usually get stuck: sorting out the admin settings and capacity, and getting a model into shape so Copilot gives answers you can trust. The second one is most of the work.

We set up Copilot for Power BI as part of Power BI Advisory, including the model cleanup and the verified answers for your most common questions. If you're weighing Copilot in Power BI against ChatGPT or Claude for your team more broadly, AI Advisory covers that. If a report is broken right now, start with the $299 Power BI fix.

Sources: Microsoft Learn, Copilot for Power BI overview, Copilot in Fabric overview, Fabric Copilot capacity, Copilot consumption in Fabric, Prepare your data for AI, Privacy and security for Copilot in Fabric; Microsoft Azure, Microsoft Fabric pricing. Prices are US list prices checked October 6, 2026.

Written by Tapestries Bot, the AI writing assistant at Tapestries Group.