Cutting an AI budget without understanding where the money went is always a bad idea, and every executive knows this in every other part of the business. The reason AI spend looks indefensible is not that AI is expensive. It is that most companies never funded an AI project at all. They opened an account and let people draw on it.
You did not fund a project. You opened an account.
Look at how the money was actually released. A large budget was allocated, and developers, or in some companies every employee, were given the ability to spend against it on tokens. No project structure, no approvals, no reporting on what any of it produced.
What happens next is entirely predictable and not remotely malicious. Every developer carries five, ten, fifty side ideas in their head at any moment. Here was free capacity and company time to try them. So they tried them.
That is not an AI cost problem. It is what happens when spending is decoupled from a deliverable, and it would happen with any resource released the same way.
MIT’s own research on enterprise AI found the same pattern at a national scale: only 40 percent of companies had purchased an official AI subscription, while more than 90 percent of employees were already using personal AI tools for work regardless. The account was open long before anyone approved a budget for it.
Your historic spend is a terrible guide to your future budget
This is the part that undermines any backward-looking calculation, and it is the strongest argument in the room.
Those budgets were set when the models were significantly weaker. Work that burned enormous volumes of tokens through repeated, frustrated re-iteration now often completes in a single session in minutes. For a finished piece of work the difference can be an order of magnitude in consumption.
So the number you are looking at describes a capability that no longer exists. Halving it is not prudence. It is applying a discount to a figure that was never a forecast in the first place.
Cutting because other people failed is not analysis
The other common trigger is what happened elsewhere. Reports of pilots that produced nothing, competitors quietly retreating, headline research on how little value has been realised. A Norstat survey of decision-makers across the Nordics found billions invested in AI with little measurable effect in Norwegian businesses, which is the sort of headline that starts these conversations.
None of that tells you anything about where your own money went. If the argument for cutting your budget is that other organisations wasted theirs, that is not a financial decision, and framing it that way is an insult to the intelligence of the people being asked to accept it. Your own failed projects are far more informative, and you already own that data.
What to do instead, and none of it is new
Put the same structure into an AI project that you put into every other project. That is the entire answer, and it is deliberately unexciting.
Establish where the money actually went, by team and by purpose, before deciding anything. Set sensible limits on consumption and say plainly that company capacity is not for personal experiments. Then ask people to come back with a proper proposal and a proper budget, tied to a deliverable, with an owner attached, and approve it the way budgets have always been approved.
What you should not do is leave the account open. An unlimited allowance handed to a group of employees with no consequences attached is not an AI strategy, and it was never going to end anywhere else.
I have watched this from the inside more than once, and the uncomfortable truth is that nothing here is about AI. It is project budgeting, applied late, and it is the same gap that leaves a pilot with a budget and a deadline and nothing else.
The question the board should ask first
Not how much are we spending, but what did the spending produce and who authorised each part of it. If you want a starting number rather than a percentage of revenue, there is a formula worth using, and it begins with headcount.
If nobody can answer that, the problem is not the size of the budget. The problem is that there was never a budget in the meaningful sense, only an account. Fix that and the number usually looks after itself.
Questions this article answers
Should we cut our AI budget?
Not before establishing where the money went and what it produced. Cutting a budget without understanding its composition is poor practice in any part of the business, and AI is not an exception.
Why did our AI spending run away from us?
Usually because capacity was released as an open account rather than allocated to projects with owners and deliverables. Spending decoupled from an outcome grows without anyone acting in bad faith.
Is past AI spend a good guide to future budget?
No. Earlier budgets were set against far weaker models, where work consumed large volumes through repeated re-iteration. The same output can now cost a fraction of that, so historic figures describe a capability that no longer exists.
Other companies are cutting AI budgets. Should we follow?
Their experience says nothing about where your money went. Your own projects, including the failed ones, are the relevant evidence and you already hold that data.
What should replace an open AI budget?
Ordinary project budgeting: a named owner, a defined deliverable, sensible consumption limits, and approval through the same process as any other investment.
If you are leading your organisation through this, this is the kind of work I do with a limited number of senior leaders each quarter. → https://anglero.com/the-reset/
Strategic AI Advisor, MerkabaPhi AS, Oslo. 450+ keynotes across 30+ countries.
Listen to the Clarity at the Top podcast on Spotify.
Enquiries: anglero.com