Power Platform Community Forum Thread Details

Copilot Studio – Issues with Numeric Condition Handling in Reference Data

Hello,

We are currently using Microsoft Copilot Studio to build an internal Welfare Points Policy Agent.
The reference data is structured in an Excel and PDF format, with each row containing four fields: Category, Points, Usage Details, and Notes. The goal is for users to ask questions like “Show me items with 120 points or less” and receive accurate results.
However, we are encountering the following issues:

Even when explicitly asking for “items with 120 points or less,” results that include items with 150 points are being returned.
The filtering accuracy for numerical conditions (such as "less than or equal to") seems to be unreliable.

We would like to request guidance and support on the following:

Improving numerical filtering accuracy: Despite providing clear conditions, the results are often incorrect. Are there any settings, recommendations, or best practices to improve this behavior?
Optimizing reference data structure: Is there a recommended format or structure for Excel or PDF files that would allow Copilot Studio to better understand and process numerical comparisons?
Limitations of Copilot Studio: Is this behavior due to known limitations of Copilot Studio or the agent’s processing capabilities? If so, are there plans for improvement?

As we are aiming for accurate and user-friendly responses through Copilot Studio, such inconsistencies significantly affect the user experience. We would appreciate any technical advice or suggestions to address this issue.

Thank you.

Quick Links

Leaderboard > Copilot Studio