Hello everybody,
I’m facing some issues with my Copilot Studio agent : it needs to analyze multiple Excel files, compare them against my criteria base (which can be a PDF), and then provide a comparison of the different responses found in those Excel files.
My agents is build with :
- No topics
- File structure is detailed in the instruction
- GPT 5
- General knwoledge on
- web research off
I have 3 excel files and 1 PDF
I’ve tried several approaches with my Copilot Studio agent :
- Using SharePoint as knowledge (Dataverse indexer)
- Uploading files manually in the conversation
- Testing different models (Chat, Reasoning)
- Trying with the code interpreter and the differents models
but the results are confusing, for example:
If I ask my agent details about a specific part of my Excel file, it either gives random numbers, approximate data, or no data at all. Even with a general comparison.
If interpretor code and GPT 5 reasoning is activated it couldn't retrieve files in its sharepoint knowledge base, I need to upload them manually but i'm facing a lot of errors like system error, it doesnt respond or is stuck in trying to analyse files.
With no interpretor and with GPT 5, it cant retrieve my PDF but only 2/3 excel files and each time it struggle to find exact data.
It is too much random to be used right now so my thought are :
AI Builder could help, but it doesn’t accept Excel files. Do I need to implement a Power Automate /agent flow process to extract Excel content and send it to AI Builder for analysis?
Should I use Azure AI? Azure search ?
I can’t analyze large Excel files with Copilot Studio, while I can do it easily and effortlessly in SharePoint with the default agent. Maybe should I stuck with sharepoint agent but lose the custom approach.
Why is there so much differences between Agents from Sharepoint / copilot studio and M365 Chat while they use the same model?
Do you have any suggestions to improve the agent in Copilot Studio for data analysis?
Thanks for your help