Ask HN: Secure use of LLMs for analyzing customer data

1 points by noname_jabroni 1 year ago | 1 comment

I would like to use ChatGPT / Claude / other "off the shelf LLMs" to analyze financial & operational data that prospects submit to my business as part of our quoting process.

We current receive data in Excel files and assign it to a team of people who take weeks to review it and prepare a proposal and estimated savings.

I believe we can do this much faster and cheaper with LLMs and that the speed trumps the accuracy (and in this case it's okay to not be 100% right)

My question: what is best practice when using ChatGPT / Claude for analyzing these files? We sign NDAs with prospects (though it doesn't specify that we would share with a 3rd party like ChatGPT). Is it enough to do a Temporary Chat?

To be clear, we have NOT done this yet - which is why I am asking. Thanks!

pfuller23 1 year ago |

generally you will find that LLMs out of the box do not do a good job on any Excel file that is big or in any way complex. your best bet would be to paste key regions of the excel into the prompt as markdown. my startup has built a solution to this, we built on some research that Microsoft published last year and have developed a way of encoding spreadsheets so that LLMs can work with them much more easily. we're early but if you want to talk you can find us on LinkedIn (https://www.linkedin.com/company/tracelight/)