How to Set Up Copilot in Power BI

Copilot in Power BI lets you describe a report in plain English and receive a working visual layout within seconds. It also writes DAX measures on request, summarizes any chart or page in narrative text, and answers ad-hoc questions about your dataset through a chat panel. Microsoft lowered the minimum Fabric capacity from F64 to F2 in April 2025, which brought the cost of entry down from over $5,000 per month to roughly $262 per month for the smallest eligible capacity. The result is that Copilot is now accessible to most organizations that already hold a Power BI license at the Premium tier.

What You Need Before Enabling Copilot

Three conditions must be met before Copilot appears in Power BI Desktop or the Power BI Service.

Capacity requirement. Your workspace must be backed by a Microsoft Fabric capacity at F2 or higher, or a Power BI Premium capacity at P1 or higher. Trial capacities do not qualify. Power BI Premium Per User (PPU) accounts can use Copilot if they also activate a Fabric trial, though trial status is time-limited.

Tenant-level toggle. A Power BI administrator must enable Copilot in the Admin Portal before it becomes available to any user in the organization. The setting is found under Admin Portal > Tenant settings > Copilot and Azure OpenAI Service. You can roll it out to the full organization or limit it to a specific security group for testing.

Workspace assignment. The workspace where you plan to build or view reports must be assigned to an eligible capacity. In the Power BI Service, go to Settings on a workspace, open Premium, and confirm the capacity assignment.

Once all three conditions are met, the Copilot pane appears in Power BI Desktop under the Home ribbon, and in the Service as a panel icon on the right side of any report.

Step 1: Enable Copilot in the Admin Portal

Sign in to app.powerbi.com as a Fabric or Power BI administrator. Navigate to the gear icon in the top-right corner and open the Admin Portal. Scroll to Tenant settings, then find the section labeled Copilot and Azure OpenAI Service.

Toggle Users can use Copilot and other features powered by Azure OpenAI to Enabled. If you want to limit the rollout, select Specific security groups and choose a group before saving. Changes propagate within a few minutes, though in some tenants it can take up to an hour.

Step 2: Assign a Workspace to Eligible Capacity

In the Power BI Service, open the workspace you want to use with Copilot. Select Settings from the top bar, then choose the Premium tab in the left panel. Under Capacity name, select your Fabric or Premium capacity from the dropdown. Save changes. Reports in this workspace can now use Copilot features.

If you do not see your capacity listed, verify that you have Capacity Admin or Contributor permissions on that capacity.

Step 3: Open the Copilot Pane in Power BI Desktop

Download or update to the current version of Power BI Desktop. In 2026, Microsoft added a dedicated Copilot entry point on the Home screen so you can start a session before opening a dataset. For report building, open a .pbix file connected to a published semantic model. On the Home ribbon, select Copilot. The pane opens on the right side of the screen.

You will see a text input field and, if you have an existing report page, Copilot will offer suggestions based on the fields in your semantic model.

Step 4: Create a Report Page with a Prompt

Type a description of what you want into the Copilot input. Specific prompts return better results than vague ones. The following patterns work consistently across most semantic models.

For a summary page: "Create a one-page summary of [table name] showing total revenue, month-over-month change, and top five customers by sales."

For a specific visual: "Add a bar chart comparing [metric] by [dimension] for the last 12 months."

For a KPI card: "Add a card showing total orders this month with a comparison to last month."

Copilot generates the page layout and places visuals using the fields it finds in your model. If a field name is ambiguous, it may ask a clarifying question or pick the closest match. You can type "change the bar chart to a line chart" or "filter to the last 30 days" as follow-up prompts.

Step 5: Write DAX with Copilot

In the Data pane, right-click the table where you want a new measure and select New measure. In the formula bar, instead of typing DAX, click the Copilot icon (the sparkle icon near the formula bar in the current version). Describe the measure in plain English. For example: "Calculate the 3-month rolling average of sales, excluding weekends." Copilot writes the DAX expression. Review it before saving, particularly the CALCULATE and FILTER logic, which can produce unexpected results if your model has bidirectional relationships.

Step 6: Generate a Narrative Summary

On any report page, open the Copilot pane and type "Summarize this page." Copilot reads the visuals on the current page and returns a paragraph-length narrative describing trends, outliers, and notable comparisons. This output is useful for embedding in email updates or executive summaries.

You can also ask targeted questions: "What drove the spike in returns in March?" or "Which product category had the lowest margin last quarter?" Copilot answers using the data in your model, not general knowledge, so the quality of the answer depends on how well your semantic model is structured and documented.

What Copilot Cannot Do

Copilot does not have access to data outside your semantic model. It cannot pull in external sources, run Python or R scripts, or create custom visuals. It also does not rewrite your entire data model or create relationships between tables. DAX measures generated by Copilot sometimes include deprecated functions or verbose logic that a skilled analyst would simplify. Always review generated formulas before deploying to production reports.

If you work with large datasets that need cleanup before analysis, VSLZ handles raw file uploads and runs end-to-end analysis from a single prompt, without requiring a pre-built semantic model or any configuration.

Practical Notes for Analysts

Give your semantic model tables and columns descriptive names before using Copilot. A column named "Rev_FY24_Q3" will confuse Copilot in ways that "Revenue Q3 FY2024" will not. Adding descriptions to measures in the model view also improves prompt accuracy, because Copilot reads metadata alongside values.

Start with read-only workspaces during initial rollout to prevent Copilot from creating report pages that overwrite existing work. You can grant Copilot full edit access once your team is comfortable with the output quality.

Microsoft tracks Copilot usage through Fabric capacity metrics. If you see capacity spikes, the Admin Portal shows per-user and per-workspace Copilot token consumption broken down by operation type, which helps identify whether specific queries or users are driving disproportionate load.