Large Excel file attached as agent knowledge, but Copilot can't reliably filter or count rows

Question

I have a Copilot agent built in Agent Builder with an Excel file attached as knowledge. It has around 30,000 rows, one row per record, with a date column and a few text and category columns.

When I ask it something that needs filtering by date range and category, like counting how many rows match a certain condition in a certain month, it usually cannot do it properly. Looking at its reasoning trace, I can see it tries several searches against the file, then just gives up and says it cannot reliably find the matching rows, so it will not give a count.

Is this a known limitation with how knowledge retrieval works for large Excel files, where it only pulls back small relevant chunks instead of being able to scan and filter the whole sheet? Is there some row count or file size where this starts to break down? And is there a better way to handle this? (I cannot break the document into smaller parts). The copilot should be able to scan excel file over 50k+ rows. Also, my excel file is just 3.3 MB. Only rows are more, file size is small only.

Would really appreciate any pointers or documentation on this. Thanks.