Q&A & Copilot (Power BI Tutorial)

Typing "total sales by city this financial year" and getting a chart back is a real time-saver for people who do not know where the fields are. Power BI has offered this for years as Q&A, and now offers it through Copilot, its generative-AI assistant. The two features overlap, their licensing is very different, and one of them is being retired, so it is worth knowing exactly where things stand.

This lesson describes Q&A and Copilot as Microsoft's documentation stated them in November 2026, how to prepare your Kesar Mart model so natural-language answers are right, and why you must still check the numbers.

Q&A: natural language with no AI licence

Q&A understands a limited English vocabulary built from your model's table, column and measure names. It appears in three places: the Q&A visual in reports (Desktop and service), the question box at the top of dashboards, and the Q&A button action. It works with a free or Pro licence; no capacity is needed.

Type a question such as:

Text

total sales by city as bar chart

Q&A picks the fields it recognises, underlines them, suggests corrections ("did you mean Total Sales?") and draws the visual. You can turn the result into a normal visual with one click. Report authors improve it with Q&A setup: adding synonyms ("shop" for Store, "turnover" for Total Sales), teaching it phrases, and reviewing questions users asked.

Copilot in Power BI

Copilot uses a large language model (through Azure OpenAI) on top of your semantic model. According to Microsoft's overview, it offers:

For readers: the Copilot pane beside a report (ask questions about the open report, get a summary), a standalone Copilot that searches across all reports and models you can access (preview), and Copilot in apps scoped to one app's content (preview).
For report authors: create or edit report pages from a prompt, summarise the model, add a narrative (summary) visual, and write or explain DAX queries in DAX query view.
For model authors: generate measure descriptions, and Copilot in web modelling (preview) that reviews the model's structure.

Prompts can be up to 10,000 characters. If you ask the same question of an unchanged model within 24 hours, you may get a cached answer.

What Copilot needs

This is where most people are surprised. Copilot is not included with a Pro or Premium Per User licence. Microsoft's requirements are:

Requirement	Detail
Capacity	A paid Fabric capacity of F2 or larger, or Power BI Premium P1 or larger. Trial capacities, free licences and PPU workspaces are not supported
Admin setting	The tenant setting "Users can use Copilot and other features powered by Azure OpenAI" must be on (it is on by default, but admins may turn it off)
Region	The capacity must be in a supported region; if the tenant is outside the US or EU data boundary, an admin must allow data to be processed outside the region
In Desktop	You must be signed in and have Admin, Member or Contributor access to a workspace on such a capacity; Copilot asks you to pick that workspace the first time

Some Copilot experiences were not yet available in every Indian region (Microsoft's list named India West among others for the standalone experience), and only English is officially supported. Copilot usage consumes capacity units, which capacity admins can monitor in the Fabric Capacity Metrics app.

For a small business with only Pro licences, that means Copilot is simply not available; Q&A was, and ordinary slicers and visuals remain the dependable route.

Preparing a model for natural language

Both Q&A and Copilot only understand your model as well as you describe it. Microsoft calls this preparing your data for AI, and the habits are good modelling anyway:

Plain-English names: Total Sales, not Amt_Sum; Stores, not dim_str.
Descriptions on tables, columns and measures. In Model view, select an object and fill in Description in the Properties pane. Copilot can draft measure descriptions for you, but read them.
Hide technical columns (keys, raw amounts) so suggestions point to the right fields.
Synonyms for the words your business uses: "outlet" and "shop" for Store, "FY" for Financial Year.
Explicit measures for anything that must be calculated a specific way, such as Margin % and Sales FYTD.
Mark the date table and relate it, so "this year" and "last month" have meaning.

A dedicated measure for the questions people ask most is often simpler than teaching the AI. For "which store is best?":

DAX

Best Store =
VAR TopStore =
    TOPN ( 1, VALUES ( Stores[City] ), [Total Sales], DESC )
RETURN
    CONCATENATEX ( TopStore, Stores[City], ", " )
        & " (" & FORMAT ( MAXX ( TopStore, [Total Sales] ), "#,##0" ) & ")"

Output

Best Store
---------------
Pune (11,341)

Always check the answer

Copilot can misunderstand "this year" (calendar or financial?), pick the wrong column with a similar name, or write a DAX query that double-counts. Treat its output like a draft from a new intern:

Read the fields it used; it usually shows them.
Cross-check one number with a measure you trust, for example Total Sales for the same filter.
Ask it to explain its DAX, then run the query yourself in DAX query view.

For example, if you ask Copilot in DAX query view for "sales by financial year" and it returns a query, compare it with one you can read line by line:

DAX

EVALUATE
SUMMARIZECOLUMNS (
    'Date'[Financial Year],
    "Sales", [Total Sales],
    "Sales FYTD", [Sales FYTD]
)
ORDER BY 'Date'[Financial Year]

Output

'Date'[Financial Year] | [Sales] | [Sales FYTD]
-----------------------+---------+-------------
FY 2025-26             | 18858   | 18858
FY 2026-27             | 20414   | 20414

If Copilot's answer for FY 2026-27 is anything other than 20,414, find out why before you share it. A common cause is a calendar-year filter instead of the April to March financial year.

Never paste confidential data into a prompt that your organisation's policy does not allow.

Microsoft is explicit that Copilot is "powered by AI, so surprises and mistakes are possible", and that preparing the model reduces, but does not remove, wrong answers.

Common mistakes

Promising the boss Copilot on a Pro licence. It needs a paid F2+ or P1+ capacity.
Building new reports on the Q&A visual. It is being retired.
Cryptic names and no descriptions. Both tools answer badly when the model is unclear.
Trusting generated DAX blindly. Read it, run it, compare it.
Assuming Copilot works in every region and language. Check the current availability list.

Key points

Q&A answers plain-English questions using your model's names and synonyms, with a free or Pro licence, but Microsoft is retiring it (February 2027 per the documentation in late 2026; earlier notices said December 2026).
Copilot provides a report pane, a standalone experience (preview), report and DAX generation, and summaries.
Copilot requires a paid Fabric F2+ or Premium P1+ capacity with Copilot enabled; Pro or PPU alone is not enough.
Prepare the model: clear names, descriptions, hidden keys, synonyms, explicit measures and a marked date table.
Check every AI answer against a trusted measure and read any generated DAX.

Exercise

Kesar Mart's owner asks "Can we get Copilot so I can just ask questions?" The company has six Pro licences and no capacity. Write a short answer (four or five sentences) explaining what is and is not possible today, and name two things Aarav should do to the model regardless of the decision.

Show solution

Copilot is not included in Pro licences; it needs the workspace to be on a paid Fabric capacity (F2 or larger) or Power BI Premium (P1 or larger), with Copilot enabled by the admin and the capacity in a supported region. A trial capacity does not qualify. Without that, the options are Q&A (still available but being retired, so not a long-term plan) and well-designed reports with slicers, cards and a drillthrough page for the common questions. If the owner wants Copilot, the next step is to compare the cost of a small F capacity on the official pricing page against the questions it would answer.

Regardless, Aarav should (1) rename and describe every table, column and measure in plain English and hide the key columns, and (2) create explicit measures such as Sales FYTD, Margin % and Best Store for the questions the owner asks most, so that any tool, human or AI, gives the same answer.

Quick Quiz

1. Which licence or capacity does Copilot in Power BI require?

2. What is the status of Power BI Q&A according to Microsoft's documentation in late 2026?

3. Which action most improves natural-language answers from Q&A or Copilot?

4. What should you do with a DAX query that Copilot writes?