AI in Finance -Fabricated Numbers wtf?
TLDR: tbh don’t read it, I’m just venting about my frustrations with using AI in finance. nothing intelligent to read here- just a rant. A lot of people I know love it, but none of them are using it for finance. This is me screaming into the void.
Over the past few months, my CEO has been having our company undergo a major AI push. Insisting that we maximize our usage of Claude as much as possible. Obviously, as a CFO, I am 100% behind maximizing the usage of any product that we are paying for. Get your monies worth! While I am a skeptic by nature, I do truly see the benefit of leveraging it.
Across all of our departments, I can see how using Claude more ultimately gives us better output, aids strategic planning, increases the team’s productivity, and blah blah blah all the great talking points about AI (for the love of God I get it, the stuff is great)
HOWEVER, for *finance specifically* I feel that I spend so much time correcting the AI, and catching its mistakes, explaining basic stuff, that I wonder if my time would be better spent doing the reports from from scratch my damn self. I won’t deny that some of the menial but tedious task that I ask the robot to do it will do perfectly fine. My frustrations stem from when I try to “maximize and leverage” my usage of Claude for more robust and time-consuming tasks only for it to be just as time-consuming for me to teach Claude to do it.
Over the last few months it has: fabricated numbers in reports that I GAVE IT THE RAW DATA FOR. It’s made erroneous, nonsensical assumptions for projections that I already gave it the assumptions for. It’s included inputs that, look good, but DID NOT ANCHOR to any source of truth provided. Only for it to apologize & admit that it’s wrong after I call it out.
Today was the straw that broke the camels back! The accountant sent over the recently filed 2025 tax returns so I use the opportunity to do some light bookkeeping, cleanup, and reconciliation to make sure everything was aligned. I uploaded multiple files for the robot to review and flag any anomalies. One of these files was the basic 2026 Jan 1st BoY balance sheet. Claude then uses it to send me this longgg analysis on why the Jan 26’ current year earnings was a red flag because it didn’t match the Dec 25’ current year earnings. Granted this is very basic and non-consequential and should not have pissed me off as much as it did.
But are you telling me that the amazing-wonderful-powerful-smarter than a human AI that everyone raves about cannot understand that a Dec 25’ balance sheet & Jan 26’ balance sheet would not have the same current year earnings-because they’re different years??? so now I have to spend my time explaining to the robot why it’s wrong and why i need it to do wtf I am paying it to do??!?
This can’t be the AI that everybody is betting their entire lives on! It can’t be, ain’t no way.
Maybe I’m using it wrong, and maybe I’m the problem. who am I kidding? Maybe I’m an idiot and I just don’t know what I’m doing. Or maybe this is some bullshit and a waste of my limited time omfg.
Obviously, I’m going to keep using it, but I would NEVER trust it to produce reports that haven’t been triple or quadrupleeee checked by ME - not another bs AI agent.
Rant over.