I’ve run into my usage cap while trying to build my own gpt. 
The challenge is in getting it to use the data provided in the files both fully and correctly. When trying out different questions it responds that it doesn’t have the data. I point it to the appropriate file and hey presto! Some times it doesn’t check all the tabs and it also fails if there’s any kind of header above the table. I’m continuously adding to the instructions (which I get gpt to provide itself). I’ve also asked it to act as an excel expert, but I’m still struggling to have it respond consistently. What am I missing? Thanks
GM, I’m no expert, but I have faced the same issues you are facing, both as a GPT and with the Assistants API.
I was trying to load my excel docs as a knowledge file, but found much better response by loading it through the chat.
Not sure if that is valid for your use case or not.
When I want to work with a particular set of data with known structure I find providing some instructions about the data set in the instructions area of the GPT is very helpful: The data may contain NaT values in Date fields, this is to be expected…This column means this… and should be used like…
Then I use very prescriptive prompts when feeding the data: Review and analyze the attached [file type] which is a [explanation of the data set] for the purpose of [use case]
If you find a better way I would love to hear about it.
I would suggest providing an example in the GPT’s instructions if you haven’t already. Something like…
For example, if a users asks for X, use the data from “Y.docx” file and provide them with the following answer “Z”.