AI for CFOs: The Four Things Finance Should Own
Finance usually gets handed AI as a cost to approve. The more useful position is four specific things finance should own outright.
Erin Moore
Fractional Chief AI Officer
A CFO should own four things in an AI programme: the true cost picture, the benefit-claim standard, the vendor concentration view, and the kill criteria. Everything else can sit with the business — those four cannot, because nobody else has both the incentive and the standing to be unpopular about them.
One: the true cost picture
Licence cost is the number everyone quotes and the smallest component. The spend that escapes the budget is usage — token or seat consumption that scales with adoption, so the better a project works the more it costs — plus integration engineering, the internal time spent reviewing output, and the migration cost when a tool is replaced.
Ask for cost per unit of work rather than annual licence. Cost per ticket resolved, per document processed, per proposal drafted. It is comparable across vendors, it exposes usage-based pricing before it surprises you, and it survives renewal. The components are broken down in AI total cost of ownership.
Two: the benefit-claim standard
Finance sets the evidentiary bar for what counts as a benefit, and this is where most AI business cases quietly fail.
Hours saved is not a benefit. It becomes one when the hours are removed from a budget, redeployed to work that has a stated value, or absorb growth that would otherwise have required hiring. Insist on which of the three, named in advance, with the person accountable for delivering it. A business case that cannot say which is a business case for a nice tool.
Ask for the counterfactual too. If throughput rose ten percent in a quarter when volume rose fifteen and two people joined, the AI contribution is not ten percent. The honest version of this is set out in how to measure AI ROI.
Three: vendor concentration
Someone has to hold the whole vendor list and see the shape of it. Six tools from four vendors, three of which resell the same underlying model, is a concentration exposure that none of the individual purchase decisions revealed.
Finance is usually the only function that sees all the invoices, which makes this unavoidably a CFO job. Pair it with the exit terms in AI contract negotiation — renewal caps and export rights are finance terms, not technical ones.
Four: kill criteria
Every funded initiative should carry, at approval, the conditions under which it stops. Finance is the right owner because everyone else becomes attached. Written at approval, kill criteria are administrative; invented later, they are a fight. The mechanics are in how to kill an AI project.
What not to own
Not tool selection, not the technical architecture, and not the standard for whether an output is good enough — that is a domain judgement belonging to the function using it. A CFO who starts adjudicating model quality has taken on something they cannot win and abandoned four things only they can do.
Three questions for the sponsor
What does this cost per unit of work at triple today's volume? Which budget line changes if it succeeds, and whose? What would we see in ninety days that would tell us to stop?
A sponsor who can answer all three has done the work. A sponsor who cannot is asking you to fund an experiment, which is sometimes the right thing to fund — but it should be labelled as one. NIST's AI Risk Management Framework and the OECD's AI Principles both put accountability before capability for the same reason.
Frequently asked questions
What should a CFO own in an AI programme? Four things: the true cost picture including usage and internal review time, the standard of evidence for benefit claims, the vendor concentration view, and the kill criteria agreed at funding.
Why is "hours saved" not a benefit? Because hours only become money when they are removed from a budget, redeployed to work with a stated value, or absorb growth that would otherwise require hiring. The business case should name which of the three, in advance.
What AI spend most often escapes the budget? Usage-based consumption that scales with adoption, integration engineering, and the internal time people spend reviewing output. Licence cost is usually the smallest line.
Should the CFO choose the AI tools? No. Tool selection and quality standards belong to the functions using them. A CFO who takes those on tends to drop the four things only finance can do.
Further reading
Tagged with:
Ready to Automate Your Business?
Let's discuss how AI automation can deliver measurable ROI for your organization in 90 days or sooner.