The AI Decision Belongs in the CFO’s Seat
Watch where the AI decision lands when an owner-managed business decides to take it seriously. It goes to the IT manager, because AI is software. Or to the keenest of the younger managers, because they seem fluent in it. Occasionally it goes to a committee with a name like "digital working group", which is where decisions go to be discussed. It almost never goes to the person who has to defend the numbers at year end.
The question nobody owns
Follow the money and the shape of the problem appears. The licences come out of one budget. The tools get chosen by whoever was keenest. The use — real or claimed — sits with a third group of people. And the question that matters, did any of this pay?, belongs to no one at all. The cost is owned. The enthusiasm is owned. The value question sits in nobody’s seat.
Joel Salinas, writing about why AI pilots fail, puts the deeper cause well: most organisations have no reliable way of turning AI excitement into accountable decisions. The popular story is that the technology isn’t ready. The truer story is upstream — nobody was ever made answerable for the outcome, so no outcome was ever required.
Notice what kind of questions an AI decision actually raises. What does this cost each time it runs? What happens to the data it touches? How would we know, in six months, whether it worked — against what? Not one of those is a technology question. They are spending, risk and evidence questions — the exact questions one seat in the business already exists to ask.
Enthusiasm gets an office
The market’s answer is to appoint somebody visibly AI. At the top end this is now nearly universal: IBM’s study of 2,000 CEOs this May found the share of large organisations with a Chief AI Officer had jumped from roughly a quarter to three-quarters in a single year.
Named ownership beats drift. A person who answers for AI is better than a technology that answers to nobody. If the choice is between a Chief AI Officer and the digital working group, take the chief.
But notice what the new seat owns — the technology, the tooling, the possibilities. It rarely owns the outcome, because the outcome is a number, and the number still has to survive the scrutiny of a different seat entirely. Enthusiasm has been given an office. The ledger stayed where it was. And in a business of your size the point is moot anyway — you are not hiring a chief anything.
The boring treatment
Treat the AI decision as what it is: a spending decision with unusual risks attached. Then give it the treatment every other serious spending decision in your business already gets. A case before the money is committed. A baseline before any claim of improvement. A review afterwards, against that baseline, with the option to stop. Controls on what it can touch while it runs. Nothing exotic — the same discipline you would apply to a new machine on the shop floor, applied to something that happens to arrive as software.
This is why I say the decision belongs in the CFO’s seat. Not because finance people understand AI better — mostly they don’t, yet — but because the habits that make AI spend honest are already the furniture of that seat. When an agent runs for £2 a time and displaces a service costing £1,000 a month, that is a finance sentence, not a technology sentence. Someone in your business should be able to say it, defend it, and notice when it stops being true.
Two things follow. First — the seat, not the person. Most businesses between £5m and £100m have no full-time CFO. The seat may be filled by a finance director, an accountant, a fractional CFO two days a month, or the owner with the spreadsheet on a Sunday. Fine. The argument is for the discipline, not the headcount. Second — the scepticism is a feature. The person least likely to be charmed by a demo is precisely the person who should decide what goes into production. Handing AI to the sceptic’s seat is not how you slow it down. It is how you make it safe to speed it up.
I should say what this doesn’t claim. Ownership does not create value — it makes the absence of value visible early, which is a different and humbler service. And the finance seat can smother as well as scrutinise; the job is to make AI spend prove itself, not to prevent it. A CFO who only says no is failing at this in the other direction.
The principle compresses to one line: put the AI decision where numbers already have to be defended. If nobody in your business currently has to defend that number, that is worth knowing before the next licence renews. And if you would like a second pair of eyes on where the seat should sit in yours — you know where to find me.