I've been on a lot of calls over the last few weeks with finance leaders debriefing on thier AI strategy and the same thing keeps coming up.
It shows up around the third month of a real AI push, once everyone's past the novelty and into the actual work. The rollout is going. Your team is using the tools, the business is using the tools, the board is happy you have a "strategy." But sometime around then, you look at your calendar and realize the efficiency gains haven't actually shown up anywhere. Your team is still tired. You're still the bottleneck. And now there's something new in the queue: validating the AI outputs that everyone else across the business is generating.
The cost of asking finance a question has dropped to zero. A product manager who used to email you with "can you pull the CAC trend for enterprise, non-competitive deals only, last four quarters?" now types that into an LLM, gets something back, and brings it to you to check. A sales leader builds a three-scenario forecast from their CRM export in fifteen minutes and asks you to confirm the assumptions. A new analyst generates a variance narrative and wants your read on whether the drivers they named are the real drivers.
This is finance being the Human API. It's not requests routing through you anymore. It's hypotheses. And hypotheses take longer to validate than raw questions take to answer.
The efficiency story assumed AI would absorb the operational work. What actually happened is that AI expanded the surface area of the work. More people asking. More models. More scenarios. More dashboards. More reports written with confidence by systems that have no idea whether the numbers they're citing reflect the real state of the business.
Finance becomes the last line of trust. Which sounds important, except it means your job is now to read and respond to everyone else's AI output instead of doing your own thinking.
The instinct at this point is to fix it with better prompts, better tools, another agent. If the AI is creating noise, use more AI to filter it. I understand the logic. It rarely works, because the problem isn't at the prompt layer. It's that the AI has no real sense of what's true about your business, and it can't, without a foundation underneath it.
The way out is slower than it looks. You have to step back and re-anchor on the handful of things that explain your business right now. Not a new forecast. Not a new planning cycle. Just the basics most finance teams stop refreshing when they're heads-down in the build:
What are the three or four drivers that actually move the numbers this quarter?
What's the current state of the business versus the narrative the board is working from?
Where's the scar tissue — the decisions already made, the bets still playing out, the assumptions that have quietly become facts?
Which workflows are load-bearing? Where do decisions really get made, not where the process map says they do?
When this is clear in your own head, AI outputs become easy to evaluate. You can see whether the hypothesis lines up with the real drivers. You can tell fast whether a scenario respects the actual constraints. You stop being the validator and start being a decision partner.
One tactical note that sounds old-fashioned but matters: go to gemba. Get out of the prompt window. Spend an hour with a sales leader, a product lead, a controller. Look at the business with your eyes and not through a model. A lot of the assumptions you're carrying into your prompts are stale, and no amount of context engineering fixes that.
This may feel like a step backward. It's not. It's the move that makes the AI usable in the first place.
What we're watching isn't an AI failure. It's the predictable result of running new tools on an old foundation. The finance leaders who come out of this cycle ahead aren't the ones with the best prompts. They're the ones who stay close enough to the business to know what the prompts are supposed to produce.
So here is the question I would leave you with.
When you close your laptop this Friday, ask yourself whether something got sharper this week because of the AI you used. Not whether you shipped more. Not whether you built something new. Whether a decision got clearer, a new picture got unlocked, or a conversation with the business got easier.
If the answer is yes, you are using it for leverage. Keep going.
If you cannot quite answer, that is worth sitting with. It's time to pause.
