Copilot read and provide answers only based on a subset of the data
Hello,
it's because chatGPT probably has some kind of RAG behind the model to extract the data and show it to the user. In Copilot Studio it's better to use another knowledge source, like Sharepoint List, Dataverse table instead of an Excel or CSV file. Why? Because that way you can filter out the data based on ODATA filter and then the model can easily show it to you. In generative orchestration mode it's also possible that the model would generate the filter by itself, based on the user prompt. Overall datasources like Excel or CSV are not recommended also from other perspectives:
1. You're not able to change/view the data that's processed by the model without downloading it separately.
2. If something changes in the data, you'd need to delete the datasource and upload it again - it indexes the file again so there's a possibility that it'll work different than before.
3. You have no control over how the file data is retrieved and presented.
So all in all - use a Sharepoint List or Dataverse table. :-)
In case of any other questions, let me know. If the answer helped you, mark it, so that others can benefit from it.
Copilot read and provide answers only based on a subset of the data
Hello,
i have the same issue. in my case we have a bigger file in the knowledge base, and copilot finds only the first 5 records. if i load the same file to ChatGPT free version, it finds all the records in the document.
It is really embarrassing, it makes copilot studio useless and way to expensive.
Under review
Thank you for your reply! To ensure a great experience for everyone, your content is awaiting approval by our Community Managers. Please check back later.
Share
Report
All responses (
Answers (
719
