Forum Discussion
Resolver IICopilot Giving Different Answers
Hi Fabric Community,
We are using Copilot (AI) in our reports now. We are seeing a difference in answers asked on the Report in the service and in the Desktop version. We ask the same question on both.
Has anyone run across the situation? If so, is there a fix or a reason for this behavior
Thanks,
DKHughes55
Hi DKHUghes55 ,
Whenever you ask a question in natural language, Copilot interprets the question and generates the required SQL or DAX query to get the answer.
The accuracy of the response mainly depends on how well we prepare and define the semantic model. For example, having clear table and column names, adding proper descriptions to columns and measures, and creating the right measures and business logic makes a big difference.
We also need to prepare the data for AI by exposing only the relevant tables and columns and providing clear AI instructions. These instructions should explain what the agent is expected to do, which columns or measures should be used for specific questions, and how the business logic works, using simple and straightforward language.
Once the semantic model and instructions are properly set up, we can either use the Fabric Data Agent directly or publish it to Copilot, where it can be used as a custom agent.
So, in simple terms, the better we prepare the semantic model and provide the right context to Copilot, the more accurate and relevant the responses will be.
Answers may vary in-terms of explanation but it should be accurate.Please mark this post as “Solution” so others can easily find it when they have a similar question. Kudos are also appreciated!
5 Replies
- tayloramy
Super User
Hi DKHUghes55,
By nature, Copilot will not give exactly the same response when the same question is asked multiple times.
Each question is individually processed by the LLM and the LLM spits out a new answer. If your model is set up properly and Copilot is able to access your data, then you should get similar answers, but you will not get word for word the same answer ever.
- FarhanJeelani
Hi DKHUghes55 ,
Whenever you ask a question in natural language, Copilot interprets the question and generates the required SQL or DAX query to get the answer.
The accuracy of the response mainly depends on how well we prepare and define the semantic model. For example, having clear table and column names, adding proper descriptions to columns and measures, and creating the right measures and business logic makes a big difference.
We also need to prepare the data for AI by exposing only the relevant tables and columns and providing clear AI instructions. These instructions should explain what the agent is expected to do, which columns or measures should be used for specific questions, and how the business logic works, using simple and straightforward language.
Once the semantic model and instructions are properly set up, we can either use the Fabric Data Agent directly or publish it to Copilot, where it can be used as a custom agent.
So, in simple terms, the better we prepare the semantic model and provide the right context to Copilot, the more accurate and relevant the responses will be.
Answers may vary in-terms of explanation but it should be accurate.Please mark this post as “Solution” so others can easily find it when they have a similar question. Kudos are also appreciated!
v-saisrao-msftCommunity Support
Hi DKHUghes55,
Have you had a chance to review the solution shared by tayloramy FarhanJeelani ? If the issue persists, feel free to reply so we can help further.
Thank you.
- v-saisrao-msft
HI DKHUghes55,
Checking in to see if your issue has been resolved. let us know if you still need any assistance.
Thank you.
The provided solution is not making sense as the underlying Semantic model is same for PBI report published as an app and used as a source for data agent. If I turn on Fabric IQ provided with Copilot on PBI Report, it is working best among all other.
