What AI actually changes in your finance function

A founder asked me recently whether AI was worth letting near the numbers. Fair question, and a common one. The honest answer is yes, with one condition that matters more than the tool itself.

AI can take days out of the mechanical part of finance work. The judgement that sits on top of it is where the value has always been, and that part does not move. Get the order right and AI makes a small finance function punch well above its size. Get it wrong and you produce more poor work, faster, and call it progress.

The number nobody mentions

The marketing around AI is loud. The results are quieter. MIT found that 95% of organisations investing in AI see no measurable return on it. Adoption has roughly doubled since 2023, yet most of that activity shows up nowhere on the bottom line.

Harvard Business Review put a name to one reason: work slop. Work that looks polished and carries nothing underneath. A tidy report, a confident summary, a clean-looking model, generated in seconds and missing the substance that makes them useful. Researchers at Stanford and BetterUp found 40% of workers had been handed work slop in a single month, and each piece cost the person receiving it on average close to two hours to sort out. The effort does not disappear. It moves downstream to whoever has to check it.

In finance, that cost is more than an inconvenience. It is a number a board acts on, or a cash call made on a model nobody pressure-tested.

Where AI genuinely earns its place

I use AI in my own work every week, so this is not theory for me.

Last month I had a complex cash modelling problem to solve. The hard part was never the building. It was working out how to approach the thing at all. I used AI to prototype options fast: three structures, then five, most of them wrong. I tested each, threw the weak ones out, and kept reframing until the right shape showed up. Hours of dead ends compressed into an afternoon. The slow, manual grind became quick and cheap, which freed me to do the part that actually mattered: the thinking about how to frame and solve the problem.

That is the pattern worth copying. AI is good at the first draft, the fast prototype, the version you react to. It clears the ground so the judgement has room.

The part that stays yours

What I never hand over is the final mile. The output check. The logic check. And the insight the model can surface but cannot weigh: this number moved, here is what it means for your cash in nine weeks.

A model that balances can still tell you the wrong story. Knowing what to verify is the skill that matters now, and it is exactly the skill a founder pointing a chatbot at their accounts does not yet have. So when someone tells me AI is risky in finance because it makes things up, I agree. That is the argument for a finance brain driving it. The danger sits in using it without the judgement to catch where it goes wrong, and it does go wrong.

Two ways to use the same tool

The same research drew a line between two kinds of user. Some people use AI to get out of doing the work. Others use it to do the work better. The output can look identical on the surface. Underneath, one is substance and one is filler.

The difference comes down to the standard you hold the work to. Used well, AI does not buy you more output. It buys you better. Sometimes that means doing less, and doing it properly.

There is a second-order effect here that growth businesses should notice. When the analysis gets cheaper, CFO-grade thinking comes within reach of a business that could never justify a full-time hire at $300,000 a year. The constraint was always the cost of senior time. That constraint has moved.

Where this leaves you

If you are weighing where AI fits in your own numbers, the question worth asking is whether it raises the quality of the decisions you make. Speed is the easy win. Quality is the one that counts. Use AI to take out the mechanical work, and keep the checking, the logic and the call on what the numbers mean firmly in human hands.

I run myCFO, a fractional CFO practice for growth businesses, and using AI well in real finance work is part of how I do it. If you would like to talk through where it could help in your business, and where it should stay out, I am happy to. No pitch, just a straight conversation. You can get in touch through the site.