Power BI Copilot and How to Turn it On

I tried to add Copilot into my Power BI workspace and ran into a wall. “Copilot isn’t available in your workspaces.” None of my workspaces had the capacity for it.

The free trial doesn’t cut it

I clicked into my profile and hit “Free trial” to kick off a Fabric trial, figuring that would unlock it. Had I read the documentation, however, I would have realized that wasn’t going to work.

It didn’t. Copilot isn’t supported in a free trial.

What kind of license do I need?

I bought the cheapest Fabric license you can buy – an F2 license. That’s about $250 a month, but you only pay for the time it’s actually provisioned (this test cost me about 10 bucks).

Once I bought the license, I tried accessing Copilot from my personal workspace and received the “Copilot isn’t supported in this workspace” message. It’s not tenant-wide- Copilot will only work in a workspace that either has dedicated Premium Capacity or Fabric Capacity assigned.

Copilot can’t see what’s not on the page

With a real workspace finally set up, I uploaded my data into Fabric as a semantic model and asked Copilot the laziest first question I could think of: “what can you tell me that’s interesting about this data?”

The answer was “nothing”. From the report, Copilot can only see what’s actually in the report, not everything sitting in the semantic model behind it.

I built a table visualization, dumped all the data into it, and tried again.

Working on this kind of thing more broadly? Here's the rest of what I've written about putting AI to work.

What Copilot actually said

That worked. With the data visible in the report, Copilot gave me a real writeup. EV registrations skew heavily toward Tesla (Model Y and Model 3 dominate), the fleet clusters hard around King and Snohomish countries (Redmond and Bellevue especially – go figure), and it caught a data issue that most of the “Electric Range” field is empty because it was likely not researched, not because the range is actually 0. Additionally, each claim had a footnote back to the report (which wasn’t that interesting, because there was only a single grid in the report anyways).

Running the same question through Claude

I uploaded the CSV file to Claude and asked it the same question. Also, a solid answer (Tesla dominates, but is softening, Rivian is punching above its weight, etc.), but all of that is coming out of a dataset with no real structure enforcing anything.

That’s the difference in my opinion. Copilot inside Power BI is working off a data model that I already built (in theory, not in this example) with relationships and filters I set up. Claude or ChatGPT is working from the raw dataset and is a lot more of a black box about what report it decides to hand you that day.

The part that actually surprised me

I asked Copilot for metrics relevant to the report. It suggested a few, then asked if I wanted them added to a new page. I said yes and the below screenshot is what it gave me.

I also went into “Editing” mode on the semantic model. Here, Copilot can actually make changes to the semantic model. In the past, I’ve always asked ChatGPT and copied and pasted stuff back and forth to come up with complicated measures, but now that can be done directly in Copilot.

That’s the real case for Copilot in Power BI over just asking Claude or ChatGPT to build a report from scratch – it’s working on top of a model you already control, instead of making its own judgement calls about what the data means.

Hope this helps!

(If you’re trying to get Copilot working reliably in Power BI and want a hand, that’s part of what we do at Tapestries Group.)