Why finance cannot push back on AI spend.
AI is creating a new category of operating expense that most organisations cannot connect to revenue, margin, productivity, or customer outcomes.
That’s the real challenge.
The problem isn’t that AI spend is growing.
The problem is that most organisations cannot confidently answer a much more important question:
Is the spend creating value?
When the answer is unclear, finance loses its ability to govern investment effectively.
And that’s where things start to break.
The moment the traditional finance playbook stopped working
Recently, Uber revealed it had exhausted its entire 2026 AI budget within the first four months of the year.
The CTO admitted that projected AI spending had been significantly underestimated.
More concerning was the acknowledgement that increasing AI usage could not always be clearly connected to customer-facing outcomes.
For a CFO, that’s the nightmare scenario:
A rapidly growing cost line.
A strong internal demand signal.
No clear framework for measuring return.
No reliable basis for approving or rejecting additional investment.
This isn’t a budgeting problem, it’s a governance problem.
Why AI is different from other technology spend
Most technology investments follow a relatively familiar pattern.
You buy software.
You hire people.
You negotiate contracts.
You forecast costs.
You measure results.
AI changes the equation: Many AI costs are consumption based and the growing pattern is most tools will be in, in the future.
The same application can cost dramatically different amounts depending on user behaviour, model selection, context length, routing decisions, adoption levels, and product design choices.
The people who influence those costs sit inside engineering and product teams, whilst the people accountable for the budget sit inside finance.
As a result, a growing number of organisations are finding themselves in an uncomfortable position:
They can see the spend and requests coming in for additional budget.
They cannot explain the economics of those investments.
And if you cannot explain the economics, governing investment becomes extremely difficult.
The AI upgrade trap
Every few months, a new generation of models arrives; More capable, more accurate and ultimately, more expensive.
Engineering teams want access immediately because the technology genuinely improves what they can build.
Finance asks a reasonable question:
What do we get for the additional spend?
That’s where many organisations get stuck. Not because the technology team lacks expertise, and not because finance is resistant to innovation. Because neither side has a shared framework for measuring whether incremental capability creates incremental value. Without that framework, investment decisions become subjective.
The conversation shifts from evidence to opinion. Opinion which ultimately, are poor basis for capital allocation.
The metric most organisations are missing
Most companies measure AI activity, very few measure AI economics.
They track:
Token consumption
• API spend
• Vendor invoices
• User activity
But these are operational metrics.
They don’t tell you whether value is being created. The organisations making better decisions are increasingly focused on a different question:
What does it cost to produce a business outcome?
Examples include:
• Cost per support ticket resolved
• Cost per customer interaction
• Cost per lead generated
• Cost per document processed
• Cost per engineering hour saved
This is where AI spending becomes governable, because the conversation moves away from technology and toward economics. Without a denominator, every cost discussion becomes subjective. With one, investment decisions become measurable.
The three categories of AI spend CFOs should separate
One of the biggest mistakes organisations make is treating all AI expenditure as a single category.
In reality, different types of spending require different governance models:
Exploratory spend: Experiments, pilots, prototypes, and proof-of-concepts. The objective is learning. Success is measured through insight and validation.
Operational spend: Recurring AI costs embedded within products or internal workflows. The objective is efficiency and scalability. Success is measured through unit economics.
Strategic spend: Major investments in new capabilities, model upgrades, proprietary systems, or competitive differentiation. The objective is enterprise value creation. Success is measured through revenue growth, margin expansion, or strategic advantage.
When these categories become blended together, organisations lose visibility into what they’re actually funding.
The result is poor decision-making rather than poor cost control.
The AI economics stack
The most effective organisations create a shared framework that connects technical decisions to financial outcomes.
A simple version looks like this:
1. AI infrastructure cost
2. Application and workflow cost
3. User adoption and utilisation
4. Business outcomes
5. Financial impact
Every layer should connect to the one above it. Most organisations can explain the first layer, very few can explain the fifth. That gap is where governance breaks down.
The missing owner problem
In many organisations:
• Engineering owns the architecture
• Product owns the feature roadmap
• Finance owns the budget
But nobody owns the economics. As AI spending grows, this becomes increasingly problematic. Because the biggest risk is no longer technical failure. It’s capital being allocated without a clear understanding of return.
Someone must ultimately be accountable for answering a simple question:
Is this investment creating measurable business value?
If nobody owns that answer, nobody owns the outcome.
Early warning signs your AI governance is failing
The symptoms tend to appear long before budgets are missed.
Watch for:
• AI spend growing faster than user adoption
• Model upgrades approved without quantified business cases
• Multiple teams consuming AI services without ownership of outcomes
• Forecasts based on vendor invoices rather than usage drivers
• No cost-per-outcome metrics
• AI costs spread across multiple cost centres with no consolidated view
These are not cost management problems, they are visibility problems. And visibility problems eventually become financial problems.
The one question every CFO should ask
For every £1 spent on AI, what business outcome improves, and how do we know?
The answer doesn’t need to be perfect, but if the organisation cannot answer the question at all, the issue isn’t overspend. It’s that nobody can explain whether value is being created. And if value cannot be measured, governance becomes impossible.
What comes next
This is the first article in a four-part series on AI cost governance for CFOs, finance leaders, and technology organisations.
Next week we’ll tackle one of the most common issues emerging across scaling businesses:
Why AI forecasting is broken, and why many organisations are underestimating future AI costs by 30% to 70%.
Because before you can govern AI spend, you first need to understand how it behaves.