The most common objection I hear from you when it comes to AI (I hear this all the time).
|
|
When I posted about Excel Agent Mode a few months back (and I spent a full day testing it before I said anything publicly, because I never recommend something I haven't tested myself).
I got push-back from financial modellers. They told me it was super dangerous and that you should never let AI decide on a multi-million project.
And for super complex models, they're right. You should not let AI make a multi-million dollar decision unsupervised.
But that does not mean you should not use it.
So today I’m going to show you how to:
- Accept that AI makes mistakes
- Not let this stop you from using it
Protecting Your Expertise
Most daily work in finance is someone from marketing asking "Can you show me what happens if I move 10K from one vendor to another?" You know the answer in 5 minutes. But you spend 55 minutes building it in Excel so it looks presentable.
You’ve spent years mastering Excel, so this is the default. But, the people who built their careers on being "the Excel guru" are often the ones that shout the loudest about AI being inaccurate.
“AI makes mistakes” sounds like you're managing risk.
But sometimes, if I'm being honest, it's people protecting their expertise.
And because these are often the most respected voices in the room (and they deserve that respect, they're brilliant), their objection carries serious weight. Leadership hears "AI is dangerous" and decides to wait.
This is the wrong decision.
Time Saved – Review Time = ROI
When I sat down with Shawn Kanungo (#1 AI & Innovation Speaker Globally) recently for my channel, he said something that stuck with me.
"Everybody wants this software to be a calculator. They want it to be deterministic. They want it to be like Excel. It's not like Excel. It's not like a calculator. It's probabilistic."
And then he followed it up with something I think every finance leader needs to hear.
"Instead of treating it like a piece of software, you have to treat it like a human. A human will make mistakes. A human will not be 100% accurate every single time. And you have to check it. And sometimes what we do as humans is we create another system that will help to check to see if that human is right."
Think about what happens when you hire a new analyst. You don't expect zero errors on their first day. You expect good work, and you build review processes around them. Sign-offs, peer reviews, reconciliation checks. That's just how finance operates.
AI needs the same treatment. Not rejection, but review.
And once you think about it this way, the ROI conversation changes completely. Instead of asking "Is AI accurate enough to trust?"), you ask "Is the time I save minus the time I spend reviewing still worth it?" For 90% of finance tasks, the answer is yes.
Still feels abstract?
Christian Martinez built an interactive brief that visualizes exactly what probabilistic vs. deterministic means, using a thought experiment your team will immediately recognize.
Check it here → Understanding GenAI Variability (2-min read)
|
|
How to Build Your Review Process
So what does this look like in practice? Here are the steps I recommend to every team I work with.
|
|
Not everything in finance carries the same weight. Low risk is your internal ad-hoc analysis, draft formatting, quick comparisons. Medium risk is management reporting, budget scenarios, variance analysis. High risk is statutory filings, board packs, audit responses. Your AI adoption approach should be different for each one.
|
|
Deploy AI on the tasks where an error is easily spotted and costs nothing. Those quick "can you build me a comparison?" requests. This is where your team builds confidence without any real risk.
And this is really important. People think they can automate everything by just throwing the problem at AI. It doesn't work like that. You can ask AI to automate the script, but you cannot hand over the full task without any checks.
|
|
I always tell people: make sure you ask AI to show you how it did something, because you don't have to make it complicated for yourself trying to audit everything manually. Get AI to help you audit it. Ask for the audit trail. Ask how it worked through the problem.
And for financial analysis specifically, don't let AI generate the answer. Instead, ask for the formula or the code, because both you can audit and both you can reuse.
You can even ask AI to verify its own output. For example:
Don't calculate it yourself. Make AI prove it to you with an extra step.
|
|
For Shaun he built an N8N workflow (N8N is a free automation tool where you can connect different AI workflows together) that sends his AI output to a second AI workflow.
That second workflow goes back to the original contract and validates what the first one produced.
But, you don't need anything that sophisticated to start. You can paste AI output into a different model and ask "audit this for errors." Two models checking each other will catch things you'd miss on your own.
(plus, if you’ve asked AI to show it’s working out, you should be able to check this too)
If you’re curious to learn more about Automation vs AI Workflows vs AI Agents you can read all about it in my previous newsletter here.
|
|
Track time saved versus errors caught. Use that data to build the case for expanding AI into medium-risk and eventually high-risk work, with the right review layers at each level.
The One Thing To Remember
AI gets you 80 to 90% of the way there. Your job is the final 10 to 15 minutes of validation, so you can then get back to higher value work.
Your financial modellers are right that AI makes mistakes. But so did every analyst in their first week, their first month, and occasionally in their first year.
But, you didn't fire them. You built review processes around them.
If we adapt our processes with AI (and not expect that AI will adapt to everything), we will be much faster at adopting it.
So stop asking "Is AI accurate enough?" and start asking "Is our review process good enough?"
That second question is one you now know how to answer 😉
Best,
Your AI Finance Expert,
Nicolas
P.S. — Hit reply and tell me: does your team have an AI review process, or is the "it makes mistakes" objection still winning? I read every reply.
P.P.S. — This came from my recent conversation with Shawn Kanungo. Watch the full video here → The Blueprint to Using AI for Finance in 2026