Edit

Share via


Enhance Q&A with Copilot for Power BI

APPLIES TO: Power BI Desktop  Power BI service

Before your business can start using Copilot capabilities in Power BI, you need to enable Copilot in Microsoft Fabric.

The Q&A visual and its natural language processing capabilities don't rely on generative AI. However, you can quickly improve the Q&A visual's ability to understand user questions by using Copilot for Power BI.

Screenshot that shows Copilot making synonym suggestions.

Prerequisites

Before you can use Copilot-generated content in Power BI Q&A, you need to satisfy the requirements for using Copilot:

Synonym suggestions

You can set up the Q&A visual to better understand natural language inputs and provide higher quality answers by adding synonyms for data in your model. By adding synonyms, you can explicitly tell Q&A what fields and tables people refer to when they use specific words or phrases, like domain-specific language.

However, adding synonyms for every data entity in your model can be time consuming, even if they're common synonyms for those names. Further, shared synonyms within an organization are only useful if the terms are defined elsewhere for the same data. Copilot provides a creative source to automatically generate synonyms, which can help to fill in the gaps left by the existing sources.

There are a few ways to add Copilot-suggested synonyms to Q&A:

  • When you first create a Q&A visual or open the Q&A setup menu, you might see a banner at the top of the visual or menu. The banner prompts you to improve your Q&A visual by getting synonyms. You can select Add synonyms to have Copilot generate synonyms for the names of your tables and columns.
  • You can also get Copilot-generated synonyms via the synonyms tab of the Q&A setup menu. Enable Copilot as a source via the Suggestion settings dropdown menu, and then select Apply. Copilot generates suggestions.
  • If Copilot is already enabled as a source, you can refresh suggestions from Copilot by selecting the Refresh button next to the Suggestion settings dropdown menu.

Whenever you get suggested synonyms from Copilot, Q&A sends your model metadata to Copilot. Your model metadata includes the names of tables and fields in your data model, but not the row contents of those tables or any user questions. Copilot returns some suggested synonyms for those names.

After you add Copilot-suggested synonyms to Q&A, they function just like any other suggested synonyms. Q&A might use them as a fallback when trying to determine what data fields a natural language input might refer to. Carefully review them in the suggestions column of the Q&A visual, remove the synonyms that are inaccurate, and approve the ones that best fit the data.

Limitations and considerations

We always work to improve the quality of the report pages, including visuals, summaries, and synonyms generated by Copilot. Here are the current limitations:

  • Synonyms generated by Copilot are subject to the same limitations as Copilot, Q&A synonyms, and model size.
  • When too many requests are sent to Copilot from various sources in a short span of time, requests might be subject to throttling. Copilot might generate suggestions for only a subset of the model.
  • Copilot might also not return results for terms for which it can't generate synonyms, or when its results are deemed inappropriate by our content filter.