What exactly is Copilot in Power BI?

Copilot is the AI assistant Microsoft has built into Power BI. You type a question or instruction in plain language, and Copilot makes a suggestion based on your reports and data models. According to Microsoft's documentation, Copilot can, among other things:

  • Create and edit report pages from a prompt, such as "make a page with revenue by month and by customer group".
  • Summarise reports, either as a whole or on a specific topic.
  • Answer questions about your data in a semantic model, in a chat.
  • Write and improve DAX, the formula language behind your measures.
  • Generate measure descriptions, so colleagues understand what a KPI calculates.
  • Add summaries to subscriptions, so a report email includes a short explanation.

There is also a separate, full-screen Copilot experience (still in preview) where you can search and ask questions across all your reports and models. Copilot also works in the Power BI mobile app.

Note: Copilot in Power BI is not the same as Microsoft 365 Copilot in Word or Outlook. They are different products with their own requirements.

What do you need to use Power BI Copilot?

This is where many SMEs get caught out. The situation as of 1 October 2026, according to Microsoft's documentation:

  • A Pro or Premium Per User licence alone is not enough. Copilot needs paid organisational capacity: Fabric capacity from F2 upwards, or Power BI Premium capacity from P1 upwards.
  • Trial capacity does not count. Copilot does not work on a Fabric trial.
  • A Fabric Copilot capacity is the route for most SMEs. You designate a capacity (at least F2) to which the Copilot usage of a group of users is billed. Those users can then also use Copilot on reports in regular Pro or PPU workspaces.
  • Your admin has to switch it on. The tenant settings include a switch for Copilot and other Azure OpenAI features.
  • The region has to be supported. For a tenant in the Netherlands or elsewhere in the EU this is normally not an issue; a handful of specific Azure regions are excluded.

As for your data, Microsoft states that it is not used to train the underlying Azure OpenAI models. Do read the current terms yourself if you work with sensitive information.

What such a capacity costs and how it compares with Pro licences is covered in Power BI licensing in 2026: Pro, Premium Per User or Fabric? In short: the smallest capacity (F2) costs a few hundred per month, on top of the Pro licences your creators and viewers need anyway.

Where is Copilot in Power BI genuinely useful?

Copilot works best as an accelerator for someone who already knows their way around Power BI. A few uses that make sense in an SME:

  • A first draft of a report page. Instead of placing ten visuals by hand, let Copilot make a suggestion and refine it. You still decide what the end result looks like.
  • DAX help. Having a measure for "revenue same period last year" or "margin excluding discounts" written out and explained saves time searching. Do check the result against a figure you know.
  • Documentation. Descriptions for your measures are dull work that often gets left. Copilot drafts them, you correct.
  • A summary for management. A short text with the monthly report that highlights the biggest shifts, so not everyone has to go through every page.

What does Copilot not do for you?

Expectations around AI are high, so it pays to be level-headed about the limits too:

  • Copilot does not fix your data. If your accounting, CRM and time tracking each use their own customer number, or the connection to your source systems is not set up properly, Copilot will give you the same wrong figures, only faster and with a convincing sentence attached.
  • Copilot does not know what you mean by "margin". Microsoft itself writes that without preparing the data model, Copilot can misinterpret the data and return generic or inaccurate answers. So you need to pin down your KPI definitions first.
  • Dutch is not officially supported. According to Microsoft, prompts in languages other than English may occasionally return relevant answers, but multilingual use is not officially supported yet. For Dutch-speaking teams, that is a practical hurdle.
  • Copilot does not replace checking. An answer from Copilot is a suggestion. For figures that drive decisions, you want a report everyone can verify.

So Copilot is no substitute for a solid data foundation. It is a layer on top that works better the better that foundation is.

An example: the managing director's margin question

A fictional example to make it concrete. An installation company with 30 employees has a Power BI report on its accounting and project system. The managing director asks Copilot: "Why did gross margin drop in the third quarter?"

Scenario 1: the data model has not been prepared. The table is called Fact_GL_Lines, three columns have "amount" in their name and nobody has defined whether margin is before or after purchase discounts. Copilot picks one, gives a plausible-sounding answer and reports a drop that is actually a booking difference. The director has no way of knowing the answer is wrong.

Scenario 2: the data model has been prepared. There is a single Gross margin % measure with a description stating exactly how it is calculated. Technical columns are hidden, tables have clear names and projects are properly linked to the accounting data. Copilot shows that the drop is mainly in two projects, and the director can click through in the report to check.

The difference is not in Copilot. It is in the work that went into the model beforehand.

How to get your Power BI reporting ready for Copilot

  1. Define your KPIs. What is revenue, what is margin, what is an active customer? Write it down and make sure there is one measure per KPI. The 10 KPIs every SME dashboard needs are a good starting point.
  2. Give tables and columns clear names. "Customer" rather than "dim_cust_v2". Copilot reads those names to understand your question.
  3. Hide what nobody needs. Technical keys, helper columns and duplicate fields make it harder for Copilot to pick the right column.
  4. Write descriptions for your measures. Let Copilot draft them and check the result.
  5. Check your connections. Copilot cannot be better than the data coming in. Are the figures right in the report without Copilot? Only then does AI on top make sense.
  6. Test with questions you know the answer to. Ask Copilot five questions whose outcome you already know. If one goes wrong, find out why before rolling it out more widely.

Is Power BI Copilot worth the money for an SME?

That mainly depends on who works with Power BI. A few rules of thumb:

  • One or two people build reports and everyone else only views? Then Copilot mainly saves some time while building. Weigh that against the monthly cost of an F2 capacity.
  • Want management to ask the data questions themselves? Then Copilot is more interesting, but only if the data model is ready for it. Otherwise management gets fast answers nobody knows are right.
  • Already have Fabric capacity for your data processing? Then the hurdle is lower, because you can use that capacity for Copilot too.

Still unsure whether your reporting belongs in Power BI at all? Read Excel or Power BI? When it is time to switch first.

Help with a Power BI setup that is ready for AI

Switching Copilot on is a tick box. A data model Copilot can do something sensible with is the real work: KPIs that are pinned down, connections that run reliably every day and a model a human can read too. See how we approach dashboarding, or read what we do with agentic AI if you want to go further than asking your report questions.