Ai is now a variable cost. Your finance function should treat it like one.
A view from the CFO's chair
For the last two years, AI sat in the budget as a tidy, predictable line. You bought the licences, you knew the annual number, and the only real decision was how many seats to fund. It behaved like software always had: a fixed cost you set once and reviewed at renewal.
That era is ending. With the move to usage-based billing for tools like Microsoft Copilot Cowork, a meaningful slice of your AI spend now behaves like a utility. It bills by consumption. The more your people use it, the more it costs, and the bill follows behaviour, not headcount.
This is not a problem to be feared. But it is a shift that lands squarely on the finance function, and the businesses that treat it as a financial discipline rather than an IT footnote will be the ones that come out ahead.
Start with the question that actually matters
Before anyone talks about credits, tokens or admin dashboards, the CFO question is the one worth asking first: what return are we buying, and can we see it?
A fixed licence cost hides this. You pay the same whether a tool transforms someone's week or sits unused, so the link between spend and value is blunt. A consumption model, for all the discomfort of an unpredictable bill, does something useful, it makes spend and value directly proportional. Every unit consumed is a unit of work that happened: a proposal drafted in minutes, an analysis completed before lunch, a multi-step project orchestrated without a person stitching it together by hand.
That is a far better basis for a business case than “we bought forty licences and hope they're being used.” For the first time, AI spend can be tied to output. The finance discipline is to make sure you're capturing that link rather than just watching the meter.
The risk is real, but it is the familiar kind
A variable cost without a ceiling is how surprise bills happen. Every CFO knows the shape of this risk, it's the same one you manage in cloud compute, in overtime, in any pay-as-you-go arrangement. The instinct when a bill spikes is to clamp down: restrict access, pull the tool, add friction.
That instinct is usually wrong. Cut people off from a tool that makes them faster and two things happen. Productivity drops, and they quietly route around you to unsanctioned alternatives, which is the worst outcome on both cost and control. The discipline is not to ban the variable cost. It's to cap it, forecast it, and make it visible, exactly as you would any other consumption line.
What good financial control looks like here
Three things turn an open meter into a managed cost, and none of them require deep technical knowledge to mandate.
A hard ceiling, set before go-live. The tools now offer genuine spend caps, not just alerts after the fact, but an actual limit on consumption. A budget alert tells you the horse has bolted; a hard cap keeps the gate shut. Insist on the cap. This single control turns “how big could the bill get?” into a question with a known answer.
A forecast grounded in real usage, not a guess. Model expected consumption before committing, set the cap and the budget against that number, and revisit it as real data comes in. A consumption cost you've forecast is a managed cost. One you haven't is an exposure.
Visibility that finance owns, not IT. The reporting now exists to see consumption in real time — by user, by team, by tool. The discipline is to make sure finance is actually looking, and looking early. Reviewed weekly in the first month, a heavy user or a misconfigured process is a small number you correct. Reviewed at quarter-end, it's a line item you explain to the board.
Put it in the business case, not the IT budget
Here is the genuine mindset shift for the finance function. AI is no longer a fixed overhead to be approved once and forgotten. It is a variable input to productivity, and variable inputs belong in the business case alongside the value they create.
That reframing changes the conversation in the boardroom. The question stops being “how many licences should we fund?” and becomes “what work are we accelerating, at what unit cost, and what's the return?” That is a question a CFO can answer with confidence, and defend. It moves AI out of the cost-centre conversation and into the investment conversation, where it belongs.
The bottom line
The shift to usage-based billing is not a reason to slow down on AI. Handled with the discipline finance already applies everywhere else, it's an upgrade: spend tied to value, cost capped by design, and a clear line from every pound spent to the productivity it bought.
The businesses that struggle will be the ones that either freeze, and forfeit the productivity, or wave it through and meet the bill by surprise. The ones that win will treat AI like any other variable cost worth having: forecast it, cap it, measure the return, and let their people get faster on terms the CFO set deliberately.
AI is going to change your cost base either way. The only choice is whether it does so on your terms.
TSG helps UK businesses adopt AI in a way that's productive, secure and financially predictable — from setting the spend controls to giving finance the visibility it needs. If you'd like to talk it through, your TSG account manager can set up the conversation.
.svg)
.svg)
.svg)
