Power BI Q&A Retirement in February 2027: What Breaks and How To Migrate to Copilot

If you search for this right now, almost every blog you find will tell you Power BI Q&A retires in December 2026. That was true when it was announced. It is not true anymore.

Microsoft has extended the retirement to February 2027, to give teams more time to work out what they depend on and move across to Copilot. Two extra months. I would not treat that as breathing room, because the work in front of most organisations is bigger than two months of effort, and almost nobody has started it.

Let me be clear about my own view of Q&A here. It was primarily demo material. Drop it on a page, type "show me sales by region", watch the room nod. Brilliant in a sales presentation. Out in the wild, with real users typing how they actually talk, it rarely landed the same way. I am sure some organisations genuinely got value out of it, but across a lot of clients over the years, the demo is what it was really good for.

Here is the thing though. This is not a feature being upgraded into a better version of itself. Q&A is being removed, and something with different requirements, different licensing and a different entry point is taking its place. Those are two very different projects.

The Q&A retirement timeline and what is actually going with this announcement

Power BI Q&A retires in February 2027, extended from the original December 2026 date. From that point Q&A stops working, the Q&A visual is removed, and reports that contain one show an error where the visual used to be.

It is worth being clear on how wide this goes, because "the Q&A visual" undersells it. Retiring at the same time:

• The Q&A visual in reports
• The Q&A button in reports
• The question box on dashboards, the "ask a question about your data" experience
• Q&A in the mobile app
• Q&A in embedded analytics
• Q&A Setup in its entirety, so synonyms, linguistic relationships, row labels, hide from Q&A, teach Q&A, review questions and suggest questions. That is Microsoft's own list, and the synonyms entry needs a caveat I come back to later, because not every synonym you have written lives in Q&A Setup

Creating new Q&A experiences stops being permitted at the same time, so anything you are still building on Q&A today is building on a dead end.

If you want it from the source rather than from me, both sit on the Power BI updates blog, the original deprecation announcement from December and the February 2027 timeline update that supersedes it on the date. Worth a read, they are short.

Key takeaways before we go further:

• Q&A retires in February 2027, and the Q&A visual will be removed rather than degraded
• Copilot is the replacement, and it needs a paid Microsoft Fabric capacity, available from F2
• Pro and PPU workspaces do not directly support Copilot, but Fabric Copilot Capacity is a route
• Embedded Q&A has no direct replacement at the moment, so if you have built Power BI into your own customer facing application, there is nothing like for like to move to
• Copilot is not available in sovereign clouds yet, meaning the isolated national and government cloud environments rather than normal Azure regions

What your users actually see on the day

This part matters more than the announcement, because it is the part that generates the support tickets.

A report page with a Q&A visual on it does not quietly lose a feature. It shows an error in place of the visual. If that page is the one a regional team opens every Monday morning, you will hear about it. Dashboards with a question box lose the question box. Mobile users lose the virtual analyst. Embedded scenarios break at whatever point the integration calls Q&A.

I get asked a version of this all the time, usually along the lines of "how bad can it really be". Fair question, and as I said above I have some sympathy with it. But it does not get you off the hook, because the error appears whether the visual was loved or barely touched. Most organisations I have worked with genuinely do not know where their Q&A visuals are, and "we will spot them when they break" is a decision to be told about it by an end user rather than finding it yourself. That is the expensive version.

Copilot is the replacement, but it is not a swap

Copilot is the direction Microsoft is pointing at, and for report consumers it is a better experience most of the time. It handles ambiguous phrasing far better than the old rule based engine ever did, it holds context across a conversation within the same session, and it is not limited to producing one visual at a time.

It is also more demanding. In my experience Q&A would limp along on a messy model, give you something, and leave you none the wiser about why it was wrong. Copilot surfaces the mess faster and at scale, which is not the same thing as being worse, but it does mean the model work is no longer optional. For a quick answer on what that work involves, a clean star schema, business friendly naming, hidden technical fields, an AI data schema, AI instructions and verified answers. I have written all of that up properly in Is Your Power BI Copilot Setup Business Ready?, so I am not going to rebuild it here.

The point for this blog is narrower. Q&A was a feature you enabled. Copilot is a capability you prepare for.

And believe me, I have sat in the meeting where someone switches Copilot on over a model nobody has tidied since 2021, asks it a perfectly reasonable question, gets a perfectly confident wrong number back, and then it is my job to explain why. Do the model work first. It is the cheaper order to do things in.

The licensing question / update everyone asked

After the original announcement this was one of the most common questions, even from people who were not using Q&A in the first place. It is the question that decides whether this whole thing is a minor inconvenience or a new line in next year's budget.

Q&A came with Pro. Copilot does not. Copilot needs a paid Fabric capacity of F2 or higher, or Power BI Premium at P1 or higher, and trial capacities are not supported. Pro and PPU workspaces do not directly support Copilot features.

So what do you do if you are a Pro shop? The route is Fabric Copilot Capacity. The capacity itself has to sit on at least an F2 or P1 SKU, and users assigned to it can then use Copilot with Power BI in workspaces running Pro, Trial, PPU, Premium capacity and Fabric capacity license modes. You assign users, you scope it with security groups, and you pay for the Copilot consumption rather than moving your whole estate onto a large capacity.

That is a genuine path and it is far cheaper than the F64 conversation people were having a year ago. Let me be honest about what it still is though. If you are a Pro only organisation, you are now buying capacity to keep a natural language capability you already had. Microsoft has made the replacement more reachable, which I am glad about, but it is not free and pretending otherwise does nobody any favours.

If the licensing side is where you are getting stuck, and it catches a lot of people out, we have written the whole thing up separately in Navigating Power BI and Microsoft Fabric Licensing. Worth reading before you commit to a capacity on the back of this deadline.

Where there is no clean answer yet

Reading through Microsoft's own posts and the Learn documentation, two situations stand out as needing a decision rather than a wait.

Embedded is the harder one, and it helps to be precise about which embedded we mean. The problem sits with embed for your customers, the PaaS pattern, where Power BI lives inside your own application and your end users never hold a Power BI licence of their own.

This is also the one place where Microsoft's position has visibly hardened, and if you only read the December post you will have the wrong picture. Back then the guidance was to use Copilot for SaaS scenarios, with PaaS support arriving in the coming months. The February update is blunter. There are now no plans to provide a Copilot replacement for embedded Q&A in PaaS scenarios at all. If you were waiting on that promise, stop waiting.

Microsoft's advice is to start evaluating migration paths well before February 2027, and it names three things to look at while being clear none of them is a direct replacement. Moving to a supported SaaS experience, if you can transition off an embedded PaaS architecture. Narrative visuals and smart narrative summaries, which give users a written explanation of what they are looking at rather than the ability to ask their own questions. And the Fabric IQ MCP Server.

That third one deserves a word, because it is not a visual you drop on a page. Fabric IQ is Microsoft's business context layer over your data, and it publishes MCP endpoints for semantic models and ontologies that an AI agent can call. So it is a route to building your own conversational experience, with your developers owning the interface, and the ontology side of it is still in preview. Genuinely interesting, and I would be watching it rather than betting a customer facing product on it this year.

Worth noting too, as far as I can see from the documentation, capacities running the Embedded license mode are not supported for Fabric Copilot Capacity, so that route looks closed off as well.

If you have shipped Q&A inside a customer facing application, that is an architecture conversation, and it needs starting now rather than later.

Sovereign clouds are the other, and by that I mean the isolated government and national cloud environments, not the commercial Azure regions most UK organisations sit in. Copilot does not support them yet, and Microsoft puts that down to GPU availability, with the advice being to review your dependencies and take it up with your Microsoft account team. So if that is you, there is no configuring your way out of this one. Plan for Q&A going and nothing directly taking its place, and see what narrative and summary visuals can cover in the meantime.

What happens to your Q&A Setup work, and where Prep data for AI fits

This is the section where the advice floating around online is, in my view, wrong, so I want to be careful and show the receipts.

The common line right now goes something like "your synonym work does not transfer, put the effort into measure descriptions instead". Sounds sensible. It is not what I found when I tested it, and the documentation is more mixed on this than that advice suggests.

Here is what I mean in plain terms. When I played with this earlier in the year, I tried to teach Copilot the words the business actually uses by writing them into descriptions. I had a measure called Sales, the business calls it Revenue, so I wrote a description along the lines of "total sales generated from revenue transactions" and then asked Copilot about revenue. It did not make the link.

The documentation does list descriptions as useful metadata, but in the detail on what Copilot is actually grounded on, descriptions are described as being used in DAX queries and Copilot search rather than as the thing that maps a business word onto a field. That matches what I saw. Where you are meant to handle that mapping now is AI instructions, which Microsoft documents as the place to tell Copilot about the terms your users will use. So keep writing descriptions, they are good practice and Microsoft has said their role will grow, just do not expect them to do the translation job on their own.

Now, the tooling being retired is Q&A Setup. The Synonyms box in the Model view Properties panel is a different surface, and that one I have tested properly. I took a measure called Sales, invented an internal term for it with no relationship to the word sales at all, and asked Copilot about it. It had no idea, and offered me suggestions instead. I then added that term as a synonym against the measure in the Properties panel, asked the same question again, and got the correct number straight back. So synonyms absolutely do carry Copilot's understanding of business language. It is the Q&A Setup route to authoring them that is going.

Here is the genuinely awkward bit though, and I have not seen anyone write about it. Both surfaces write to the same underlying model linguistic schema, and the documentation lists synonyms from that schema among what Copilot is grounded on. The linguistic schema is the thing Q&A Setup authors. Which is being retired. On top of that, if the tabs in Prep data for AI are greyed out, the documented fix is to enable Q&A for your model, so today the replacement tooling leans on the feature being removed. Microsoft flagged back in the December announcement that a separate file format is coming for Copilot to break that dependency. Announced, not shipped. My expectation is that the Properties panel box survives and synonyms simply move house, but that file format announcement is the one to watch.

So what do you actually do with this? My position, and I will happily revise it if Microsoft ships something that changes it:

• Stop investing effort in Q&A Setup depth. Teach Q&A, linguistic relationships and phrasings are going and none of that is coming back
• Keep using the Synonyms box in the Model view Properties panel. It is not the Q&A Setup tooling, and it does feed Copilot today
• Move business terminology into AI instructions, which is documented as the place to handle mappings between the language users use and the fields in your model
• Keep adding descriptions, for readability now and for Copilot later
• Test your own models rather than trusting either me or a blog you found on page one, because this area is moving

How to audit your estate before the date

This is the bit I would be doing in the next few weeks, not in January. Microsoft's own recommended next steps open with exactly this, inventory your Q&A dependencies across reports, dashboards, mobile, embedded solutions and Q&A Setup artifacts, so this is not me inventing work for the sake of it.

Find everything first. Q&A visuals and Q&A buttons across reports, question boxes on dashboards, mobile usage, and any embedded integration that calls Q&A. The scanner APIs and your usage metrics will get you most of the way there, and this is exactly the sort of thing a decent governance process should already be able to answer. If it cannot, that is worth noticing in itself.

Then work out who uses them and how often, because that decides the response per report. Some will be replaced with a Copilot experience. Some will be replaced with a couple of well designed visuals, which is honestly the right answer more often than people expect. Some can simply be retired, and nobody will miss them.

Then, and only then, sequence the model work and the licensing decision. Not every model that has a Q&A visual on it deserves to become a Copilot ready model, and that is a call worth making deliberately rather than by default. Again, the depth on the model side is in the Copilot readiness blog.

Most organisations do not know how many Q&A visuals and dashboard question boxes they have, and this deadline forces a tenant wide answer to a question nobody has been asked before. At Metis BI this is a lot of what we do, finding what is actually out there, deciding replace or retire report by report, working out the licensing route that fits the estate rather than the one that sounds impressive, and then preparing the semantic models that will carry the Copilot experience properly. It suits organisations with a real Power BI footprint, a February deadline and no appetite for finding out the hard way. Learn more about our Copilot readiness work, and if the bigger issue is that you do not know what is in your tenant, that is a governance conversation.

Summary

Q&A retires in February 2027. The visual gets removed, the dashboard question box goes, embedded breaks, and all of Q&A Setup goes with it. Copilot is the replacement and it needs paid capacity, with Fabric Copilot Capacity from F2 as the sensible route for Pro and PPU estates. Embedded PaaS and sovereign cloud users need a plan rather than a migration.

The extension to February bought you two months. What it did not do is make this smaller. Find what you have, decide what deserves replacing, and treat the model work as the project it actually is.

Q&A is not being upgraded. It is being removed, and something more capable and more demanding is taking its place.