Microsoft’s AI assistant, Copilot, aims to speed up both report development and report reading in Power BI. So what can Copilot actually do, what do you need to use it, and what should you watch out for?
What can Copilot do?
- Report summaries: It can summarise the key findings on a report page in text.
- DAX support: It can suggest a measure such as “sales change versus last year”, together with an explanation.
- Visual and page suggestions: It can suggest visuals or draft a report page based on needs described in natural language.
- Agent experiences: With the agent experiences announced as generally available in October 2026, it gains the ability to create and edit semantic models and reports from requirements or a screenshot.
What do you need to use it?
- Paid capacity: Copilot requires the report’s workspace to be assigned to a paid Fabric capacity of F2 or above. Trial capacities and shared capacity are not supported, and Copilot is disabled when the capacity is paused.
- Admin setting: Copilot is turned off by default. It must be enabled in the tenant settings of the admin portal. Initially, it is recommended to enable it only for a specific security group.
- Region support: The capacity must be in an Azure region where Copilot is supported.
Things to watch out for
- Data model quality is decisive: Copilot produces much better results in a model with meaningful table and column names, descriptions and correct relationships. In a messy model, its suggestions are messy too.
- Validate the outputs: Always compare suggested DAX measures and summaries with the source figures; AI output is not always correct.
- Security is preserved: Copilot respects row-level security and sensitivity labels; users only see data they have permission to access.
- Cost: Copilot usage consumes capacity. With heavy use, you may need to review your capacity planning.
Conclusion
Copilot makes a well-built Power BI environment much more efficient, but it does not fix a weak data model. The best way to prepare for Copilot is to first simplify the semantic model and clarify measure definitions. Our article on common DAX mistakes can also help with this.