How to transform data with Copilot in Dataflow Gen2

João Barros 06 de July de 2026 4 min read

Copilot in Fabric brings AI assistance to Dataflow Gen2, letting you transform data with Copilot using natural language, without having to master Power Query. Instead of hunting for the right button or writing M code by hand, you describe what you want — filter rows, create a column, or group values — and Copilot generates the steps for you. It is a fast way to clean and prepare data, and to learn Power Query while you work.

Prerequisites

A paid Fabric capacity (F2 or higher) or a free Fabric trial.
Copilot enabled by your administrator in the tenant settings.
A workspace assigned to that capacity, with permissions to create items.
A simple data source to practice with, such as a sales CSV file.

Step 1: Create a Dataflow Gen2

Open your workspace in Fabric, select + New item, and choose Dataflow Gen2. Give it a clear name, such as Clean_Sales, so you can find it easily later. The Power Query editor opens — the same visual environment already found in Excel and Power BI — where you will prepare the data.

Step 2: Connect to a data source

Click Get data and choose the right connector — for example, Text/CSV for a file. Enter the path or upload the file and confirm. In the central area you will see a preview of the first rows, which is the basis Copilot will work on.

Step 3: Open the Copilot pane

On the Home ribbon, click the Copilot button. A pane opens on the right with a text box where you type your requests. If the button is greyed out, Copilot is not yet active on your capacity; in that case, ask your administrator to enable it.

Tip: make one request at a time. That way it is easier to review each step and fix things if something goes wrong.

Step 4: Ask for a transformation in natural language

Write a clear, specific request. For example, to clean the sales table:

Keep only the rows where Country is "Portugal" and remove the Phone column

Copilot interprets the request, applies the transformation, and records it under Applied steps on the right. You can keep chatting and chain new instructions on top of the previous result.

Step 5: Create a custom column

One of the most useful capabilities is generating formulas from a description. Ask, for example:

Create a column called Margin equal to (Revenue - Cost) divided by Revenue

Copilot writes the formula in M and adds the column. Always review the result — an example of the code it might generate:

= Table.AddColumn(Source, "Margin", each ([Revenue] - [Cost]) / [Revenue])

Step 6: Explain a step with Copilot

If you inherited a dataflow from someone else or do not understand a step, use the explain feature. Type in the pane:

Explain what the last step does

Copilot describes, in plain text, what the M code is doing. It is a great help for learning and for documenting the process for your colleagues.

Check the result

Confirm three things: that each request created a step under Applied steps, that the preview shows the expected data (filtered rows and the new column filled in), and that the M formula does what you asked. This review matters, because Copilot can misread an ambiguous request. When everything is correct, set a data destination (for example, a Lakehouse or a Warehouse) and click Publish to save and run the dataflow.

Conclusion

You have just transformed data with Copilot in Dataflow Gen2, saving time and without writing Power Query by hand. The next step is to try richer requests — group by category, convert data types, or merge two tables — always validating the generated code. What will be the first transformation you ask Copilot to do on your own data?