The Board Wants AI. Your Books Aren't Ready.

A CEO forwarded me a note from his last board meeting. One line, from a new director: "Why isn't finance using AI yet?" He wanted help saying yes before the next meeting. I told him the yes was the easy part.

He isn't alone in feeling the push. Bottomline surveyed 414 CFOs across the US and UK this year. 90% said they were under board pressure to adopt AI in finance. 76% said they were being pushed to move faster than their data, systems, and controls could support. And 79% said their own finance data wasn't clean, connected, or controlled enough to trust as a foundation for AI at all.

Read those three numbers together. The board wants it. Finance knows the plumbing can't hold it. That gap between the two lands on one desk.

The board isn't asking for a tool

When a director asks about AI, the reflex at a growing company is to go buy one. Pick a vendor, run a pilot, put a slide in the next deck. That answers the question as asked and misses what the board actually wants.

They want fewer surprises. Faster answers when they ask one. A forecast that holds for more than a week. A director who says "use AI" is really saying "I don't want to hear next quarter that a number moved and nobody saw it coming." Bolt an agent onto a ledger with double-counted transfers and miscoded revenue and you don't deliver any of that. You deliver a confident wrong answer, produced faster, with a logo on it.

That same survey found fewer than half of these CFOs felt confident forecasting cash even 30 days out. Automating on top of that doesn't fix it. It hides it.

Name the sequence, not the tool

The stronger move in that boardroom is to reframe the question. Not "yes, we're piloting something," but "here's what has to be true before AI earns its keep, and here's where we are on it."

I've walked a founder through exactly this. Reconcile every bank and card feed monthly, so the inputs tie before anything reads them. Settle on one definition of revenue that matches across billing and the ledger, so an agent isn't choosing between two. Write down approval thresholds, so a machine has an actual rule to enforce instead of a vibe. None of it is glamorous. All of it is the difference between AI as a multiplier and AI as a faster way to be wrong.

A board that asks for AI is asking for confidence. The fastest way to lose theirs is to give them an automated number built on books you already don't trust.

Done in that order, the pilot lands on something solid, and the board conversation shifts from "why not yet" to "here's what we turned on and why we trust it." That's a better slide anyway.

That director asked a technology question. The honest answer is an operations one. Giving it, instead of buying your way past it, is the most useful thing a finance leader can do this year.