
How to set your 2027 AI budget: A CFO’s guide built on European spend data
.jpg)

Most 2026 AI budgets were set from a list of tools, and most came in high. This guide shows CFOs how to set a 2027 AI budget from actual spend across every entity, card, invoice and tool, using European spend data and a survey of 28 finance leaders.
- What is an AI budget, and why doesn’t it behave like a software budget?
- Why AI budgets overran in 2026
- How to set your 2027 AI budget in six steps
- Revisit vendor assumptions before you lock the 2027 line
- How to keep your AI budget on track through 2027
- AI budget FAQs
- Set your 2027 AI budget from real numbers
By submitting this form, you agree to receive emails about our products and services per our Privacy Policy.
An AI budget is the planned annual spend on AI tools, models and usage-based services, set as its own line rather than folded into the software budget.
Most 2026 AI budgets were set from a list of tools, and most came in high. The extra spend rarely came from tools finance had never heard of. It grew inside the ones finance had already approved. The fix for 2027 is to build the line from actual spend across every entity, every payment route and every tool. Then, control it at budget, policy, approval and payment level.
We tracked the AI spend of 1,426 European companies across 20 countries through the first half of 2026, using our own card and payment data. Two findings from The Latest State of European AI Spend report matter most if you’re setting a 2027 budget. Among companies already buying AI, spend rose 478% between H1 2025 and H1 2026. And in a survey we ran alongside it, 71% of the European finance leaders said AI came in over budget.
This guide covers why AI behaves differently from the rest of the software budget, six steps to set the 2027 line, and how to keep it on track once the year starts.
What is an AI budget, and why doesn’t it behave like a software budget?
An AI budget covers everything a company pays for AI: subscriptions, model and API usage, and the smaller tools teams buy for themselves. It behaves differently from the rest of the software budget because the cost driver is usage, not seat count. This means the bill can rise without a single new licence being bought.
Seat-based vs usage-based AI pricing
There are three kinds of AI spend. Seat-based subscriptions charge a fixed price per user and behave like any other software contract. Usage-based models and application programming interfaces (APIs) charge for tokens, calls and credits, so the bill rises with use rather than headcount. Small tools bought by individual teams sit somewhere between the two and change month to month.
Payment route matters as much as pricing. Team-level tools are usually paid on company cards, while enterprise seats and API commitments arrive as supplier invoices. An AI budget has to cover both routes. If it doesn’t, the invoiced part gets booked under general software and disappears from the AI line entirely.
| Spend type | What drives the bill | How it’s usually paid | How to budget it |
|---|---|---|---|
| Seat-based subscriptions | Number of licensed users | Company card, or supplier invoice for larger contracts | Price per seat against planned headcount, reviewed at renewal |
| Usage-based models and APIs | Tokens, API calls and credits consumed | Supplier invoice for committed contracts, card for pay-as-you-go | A range built from the last 12 months of consumption, reforecast monthly |
| Tools bought by individual teams | Teams adding tools as they find use cases | Company card, sometimes expense claims | A capped allowance per team with a named owner, reviewed quarterly |
Why AI is now a fifth of the software bill
AI took 19%* of software spend across Payhawk’s European customer base in H1 2026. Usage expands inside tools once a team finds a use case, so a flat seat count can still produce a bigger bill. General-purpose models keep scaling as a company grows; narrower single-purpose tools tend to plateau early.
Gartner forecasts worldwide AI spending will grow 49.5% in 2026 to around $2.67 trillion, reaching about $3.64 trillion in 2027. AI software alone is predicted to rise from $461.6 billion to $656.4 billion. A line growing at that rate needs its own budget, its own owner and a monthly look, not an annual renewal check.
*Source: Payhawk card and payment data, H1 2025 to H1 2026. Large invoiced enterprise and API contracts, Microsoft Copilot and Google Gemini are excluded.
Is your AI line growing faster than your plan?
See where European companies’ AI budgets go, which tools are growing fastest, and what finance leaders plan to change in 2027.
Why AI budgets overran in 2026
AI budgets overran in 2026 because spend grew inside tools finance had already approved, faster than an annual plan could absorb. Budgets were beaten even at companies whose finance leaders could name every AI tool they pay for. In our survey, 89%* of the leaders expect AI spend to grow again over the next 12 months. None expect it to fall.
*Source: Payhawk survey of 28 European finance leaders, summer 2026. Figures are self-reported.
Knowing your AI tools isn’t knowing your AI bill
Most of the group could account for their tools: nearly 54% of the finance leaders we surveyed said they know exactly how many AI tools their organisation pays for. Another 36% have a rough idea. Overruns were still common across the group. 35% of the ones who went over said the gap was large rather than marginal. Tracking which tools a company pays for is a different job from tracking how fast each one is growing.
Where AI spend hides: cards, invoices, entities and expense claims
The median company we tracked pays three AI vendors. That spend is spread across several entities, many company cards, supplier invoices and a handful of expense claims. No single statement shows the total. So, finance sees the real number at month-end or after a renewal, when the money is already committed. A small amount of shadow AI, bought on personal cards and expensed, widens the gap further.
Daniel Todinov, VP Finance & FP&A at Payhawk, knows the problem from the inside:
Many of the finance teams I speak to can name every AI tool they pay for. What catches them out is the timing: the card feeds land this month, the invoices land next quarter, and by the time both are in, the budget has already gone. The teams that stay on budget are the ones watching both routes in one place, well before the quarter closes.
In Payhawk, card spend counts against the budget once it’s approved, and AI contracts raised as purchase orders count as committed spend before the invoice arrives.
More companies are buying AI every quarter. The Office for National Statistics (ONS) found that 30% of UK businesses used at least one AI technology in September 2026, rising to 51% of businesses with 250 or more employees. Adoption rose from 56% to 77% across Payhawk’s European customer base between January 2025 and June 2026.
How to set your 2027 AI budget in six steps
Setting an AI budget for 2027 should start from what the company spent, splitting it by how it behaves. It needs to give every line an owner with controls attached:
1. Build a baseline from actual spend, not a tool list
Pull the last 12 months of AI spend from every entity, card, supplier invoice and expense claim into one view. A tool list tells you what finance has approved. It won’t tell you what those tools cost last month. Tag the spend with a dedicated AI expense category or a custom field so the number is reportable without someone rebuilding it each time. Tracking AI spend in Payhawk walks through both options. For groups, multi-entity spend visibility is what turns five entity figures into one baseline. Expect that baseline to land above the budget you set for 2026.
Who owns it: Financial controller
2. Split the AI line by spend type
Divide the baseline into committed subscriptions, experimentation and trials, and usage-based consumption. Each needs a different owner and a different rule. Committed subscriptions get an annual decision at renewal, trials get a cap and an expiry date, and consumption gets a monthly review. Paying each committed AI tool from its own subscription card and running trials on single-use cards keeps the two apart in reporting from the moment of payment, rather than at reconciliation.
Who owns it: Financial controller, with FP&A
3. Model growth as a range
Build three scenarios: flat usage, planned growth and high consumption. Anchor the growth case on how fast spend grew inside existing tools rather than on headcount. AI spend rose 478% among companies already using AI between H1 2025 and H1 2026, which is the pattern a 2027 budget has to survive. A range gives the board a figure to approve and gives finance a trigger that says when to act.
Source: Payhawk card and payment data, H1 2025 to H1 2026. Large invoiced enterprise and API contracts, Microsoft Copilot and Google Gemini are excluded.
Who owns it: FP&A manager
4. Benchmark against companies your size
AI spend per employee only means something against companies of a similar size. The median across Payhawk’s European customers was €45 per employee in H1 2026, and the figure varies more than tenfold between the smallest and largest companies in the data. A single European median is too wide a range to plan against, so compare against the band your company sits in.

Our report breaks AI spend down by company size and vendor, so you can benchmark your AI spend against companies like yours.
Who owns it: CFO, with FP&A
5. Flag duplicate tools before renewal
With a median of three AI vendors per company, overlap is common: a coding assistant, a meeting note taker and a general model can all end up transcribing or drafting the same thing. Check which tools do the same job before each renewal date rather than after it. Give procurement the renewal calendar early enough to negotiate. The subscription and SaaS spend audit playbook sets out the sequence, and the same review flags price rises and vendors being paid twice. Seat usage still has to come from each vendor’s admin console.
Who owns it: Procurement, with IT
6. Give every AI line an owner and layered controls
Control over AI spend isn’t one switch on a card. It sits at four levels, and a budget holds only when all four agree. One finance owner holds the total. The chief information officer (CIO) or head of IT co-owns which AI tools get approved, because IT owns tech-stack direction. Layered spend controls is the reference for how the levels fit together.
The four levels of AI spend control
| Level | What it controls | Payhawk feature |
|---|---|---|
| Budget | Who owns each AI line and how much of it is used, visible in real time | Budgets, with entities as dimensions for groups |
| Policy | Which vendors and spend types are allowed, across groups of cards | Spend policies, including blocking unapproved or duplicate AI vendors |
| Approval | Who signs off, by spend type and amount, before and after money is committed | Before: card requests, fund requests, and purchase requests and purchase orders for invoiced AI contracts. After: multi-level approval of card payments, bills and reimbursements |
| Payment | What the card itself will let through at the point of purchase | Card limits, advanced card controls, and card reassignment when someone leaves |
Leon Steenbrink, CFO at Mercell Nederland, replaced a single approval level with several:
My first reaction to Payhawk was that I liked the opportunity to split the cards in virtual and physical cards and assign different responsibilities and approval levels to various persons. We had just one level and that wasn’t enough for me to keep control.
Purchase order approval matters more for AI than for most categories. It counts an invoiced API commitment against the budget when it is committed rather than when it is paid, often two quarters later.
Who owns it: CFO, with IT and procurement
How does your spend compare with companies like yours?
Your AI line sits inside your software bill. See where your software and card spend stand against companies of your size and industry.
Revisit vendor assumptions before you lock the 2027 line
OpenAI took 83% of AI card spend in H1 2025. A year later Anthropic held 63%, OpenAI had dropped to 16%, and Cursor sat just behind at 13%. The Latest State of European AI Spend report tracks the whole swing month by month. These are shares of spend on company cards within our customer base, not the European AI market, and large invoiced enterprise and API contracts sit mostly outside the data.
The budgeting point is simpler than the market one. Rates, volume commitments and vendor assumptions set more than a few quarters ago are worth re-checking against current usage before they roll forward into the 2027 line.

How to keep your AI budget on track through 2027
An AI budget is a monthly process, not an annual number. The owner of each line needs to see utilisation as spend happens, not read about it in a month-end report three weeks later.
Track budget vs actual in real time
Give every AI line a named budget owner who can see utilisation against plan at any point in the month. Real-time budget tracking counts card payments, reimbursements and supplier invoices against the budget as they’re approved, and shows approved purchase orders as committed spend before the invoice arrives, so an invoiced API contract and a team’s card spend show up against the same budget. Multi-entity groups can hold one group budget with entities as dimensions, and import budgets from your FP&A tool rather than rebuilding them.
Todinov runs the same rule on Payhawk’s own spend:
Payhawk gives me real-time spend transparency, so I can focus on advising the business rather than searching for numbers.
Reforecast monthly, not annually
Set triggers that force a reforecast rather than waiting for the quarter to close: a pilot moving into production, a team’s consumption jumping two months running, or a new entity adopting a tool the group already pays for. Pair that with a renewal check before auto-renewals land. Route larger AI commitments through purchase requests, so they reach the right signatory before anything is signed or charged, rather than clearing on a card limit. Spend policies for AI tools keep the rules applied across every card rather than one at a time. Reassigning cards when an employee leaves keeps a subscription running under a named owner, instead of breaking or renewing with nobody watching it.
Book a demo to see how Payhawk brings AI spend from every entity, card, invoice and tool in a single view.
AI budget FAQs
How much should a company budget for AI per employee?
Median AI spend was €45 per employee across Payhawk’s European customers in H1 2026, and it varies more than tenfold by company size, so use the band your company sits in rather than the median. The European AI spend report gives the breakdown. A broader US measure from the Federal Reserve Bank of Atlanta put average spend at $1,358 per employee in 2025, though that includes hardware, training and IT support.
Should AI be a separate line from the software budget?
Yes, as its own line. Software budgets assume the bill follows seat count, and AI bills follow usage. Keeping AI inside the general software line hides growth that happens without a single new licence being bought, which is the pattern behind most 2026 overruns.
Who should own the AI budget?
One finance owner holds the total, usually the CFO or finance director, with a named budget owner for each team or tool below that. IT co-owns which AI tools are approved, because tech-stack decisions sit there, and procurement owns renewals and vendor terms.
How often should finance review AI spend?
Monthly for usage-based tools and APIs, where consumption can change the bill without any new purchase. Fixed subscriptions can be reviewed at renewal, provided the renewal date is in a calendar someone watches. Budget owners should see utilisation continuously rather than waiting for either review.
Set your 2027 AI budget from real numbers
A 2027 AI budget built from a tool list will miss. One built from actual spend across every entity, card, invoice and tool, benchmarked against companies the same size and controlled at budget, policy, approval and payment level, will hold. Start with the numbers: benchmark your AI spend against your peers, or book a demo to see AI spend from every entity and payment route in one view.
Sources
- Payhawk card and payment data, H1 2025 to H1 2026, covering 1,426 European companies across 20 countries. Large invoiced enterprise and API contracts, Microsoft Copilot and Google Gemini are excluded.
- Payhawk survey of 28 European finance leaders, summer 2026. Figures are self-reported.
- Gartner, worldwide AI spending forecast for 2026 and 2027.
- Office for National Statistics, Business insights and impact on the UK economy, 8 October 2026.
- Federal Reserve Bank of Atlanta, How much are firms spending on AI (and what will happen to headcounts)?, 6 May 2026.
Zhenya is a creative content strategist specialising in crafting SEO-driven narratives across tech, SaaS, and B2B. At Payhawk, she blends storytelling, data, and product insight to create content that helps finance teams and drives measurable impact.
Related Articles


Spend management for SaaS: How to get control of your rising software costs

.jpg)
