European Spend Report 2026: €7B Lost Annually to SaaS and AI Overspend

Before we get into the how and why, here's what studying 2,500 European SMBs’ spend over the last 12 months showed us:

Key Takeaways

🇪🇺 🇬🇧 European level*:

• Vendor-level price variance points to a market-wide opportunity of roughly €7 billion a year.
• Tool de-duplication alone could free up €3 billion a year in the market.

🔍 Report level:

• Vendor-level price variance across European SMBs amounts to €69–76 million a year in recoverable spend: proof that even identical software, bought by companies the same size, rarely costs the same twice.
• Tool de-duplication alone could free up €15M if teams had more visibility and control over how that spend takes place.
• More than two-thirds of companies using AI tools already pay for two or more competing products at once — usually without ever having decided to.
• Total AI spend grew 340% in twelve months, even as the number of companies buying it barely moved. The cost is compounding inside firms that have already converted, not spreading to new ones; talk of AI’s “democratisation” hasn’t kept pace.
• The typical AI bill swings by 61% from one month to the next, and its priciest months can run up to eight times its cheapest — a pattern no fixed budget line was built to survive.

*For modelling details, see Sample & Method section.

Download in full size

Every CFO has a theory about SaaS sprawl. None of them have seen the actual numbers. At Spendesk, this is what we do every single day.

That's the vantage point you get running a spend management platform that processes more than €10 billion in B2B payments across Europe: we don't see what companies say they spend on software, we see what they actually pay, month after month, tool by tool. This report is built on twelve months of real transaction data — invoices, subscriptions, card payments — from 2,500 carefully chosen and anonymised European companies across nine major industries and four size bands.

The hidden cost of European SaaS sprawl: Why software waste is structural

European companies now spend close to €250,000 a year on software, on average. That number has been climbing for years, and almost nobody outside finance has been asked to explain why.

Part of the answer is simple neglect. External benchmarks (Zylo's 2024 SaaS Management Index) put average licence utilisation at 49%. Half the seats a typical company pays for go unused, month after month, because nobody audited them, reassigned them, or thought to check.

The other part is pricing. The same software, sold to companies the same size, does not cost the same amount twice. Modelled across every European company with 50 to 249 employees, that gap is worth something in the order of €7 billion a year. This is the scale-up danger zone: headcount grows faster than procurement discipline does. A company can run 30-plus SaaS tools before it has anyone whose job is negotiating the rate on a single one of them.

Duplication adds to the bill on top of that. Most companies using AI tools now pay for two or more competing products at once. So do most companies using design software. The pattern holds category after category: running three overlapping tools has stopped being a mistake and started being the default.

On an invoice, it just looks like an ordinary line item from a vendor nobody double-checked.

The waste, in other words, is structural.

Why AI changes the control problem: Usage-based bills can grow faster than budgets

What makes this moment different from previous SaaS sprawl cycles is AI, and the shift is showing up fast.

A year ago, OpenAI was the clear leader in AI vendor spend across European companies, ahead of everything else by a wide margin. By early 2026, Anthropic had overtaken it: spend on Anthropic alone grew more than sevenfold in the second half of the twelve-month window, while OpenAI's own spend plateaued, then eased back. Cursor's bill more than doubled over the same period. Dust, Lovable and Mistral barely register yet, but that won't stay true for long at this rate.

Traditional SaaS bills don't move like this. A €500-a-month Figma invoice is still €500 next month. AI billing scales with every prompt, every agent, every automated workflow, which means the bill itself becomes unpredictable in a way finance teams have never had to plan around. There's no renewal date to trigger a review and no budget line built to absorb a swing of several times its usual size. The spend accumulates until someone happens to notice, and by then it has already compounded.

How European finance teams can reduce SaaS waste and control AI

The window to get ahead of this is closing, and the advantage compounds for whoever moves first.

Benchmarking

The first problem is a lack of benchmarks. Finance teams can't negotiate when they don't know what other companies are paying, and SaaS vendors are notoriously opaque about pricing. Visibility into peer spend is the starting point, not a nice-to-have.

Usage monitoring

The second problem is spend efficiency. As AI moves from experiment to infrastructure, cost-per-output becomes the metric that separates companies scaling AI profitably from companies just scaling AI spend. The tools that solved SaaS sprawl in the 2010s are now being asked to do something harder: move from tracking licences to tracking consumption. Companies that treat AI spend as a strategic input, not a line-item afterthought, will spend the next twelve months pulling ahead of the competition.

Real-time controls

This report covers three areas finance teams can find real savings. One thing they have in common is that real-time controls on spend - automated limits that enforce policy at the point of spend, not after it - could significantly decrease wastage across the board.

Almost every finance team we talk to believes their company already runs about as efficiently as it can. The data below says otherwise.

1. Price variance: Same tool, same size team, wildly different bills

Key takeaway: European companies could be overpaying €7 billion a year for SaaS

Spendesk compared per-employee monthly spend across common per-seat SaaS vendors, while controlling for company size. The analysis found approximately €69 million in annual overpayment for the 2,500 companies studied, with businesses often paying two to four times more than the most disciplined buyers for the same software.

Software vendors publish prices. What companies actually pay doesn't follow them. Across more than 35 of the most common per-seat SaaS tools, we measured how much each company pays per employee per month for the exact same product, then compared every company to the most disciplined quartile of buyers at its own scale. What's left over after controlling for headcount is the overpayment: roughly €69M a year, or about €40K per company that overpays.

The size-adjustment is the point here. The obvious objection is that bigger companies simply pay more — and they do; a 200-person company will always spend more on Slack than a 20-person one. That's exactly what we corrected for. Benchmark on spend per employee, and the variance barely narrows. Atlassian shows a €2 vs €12 per-head gap. Notion, €2 vs €7. HubSpot, €6 vs €24. Across the board: a three-to-six-fold per-head gap that headcount can't explain.

This isn't portfolio mix. Wide spreads on a vendor like Adobe or Atlassian can be partly explained by product breadth but when a single-product tool like Slack shows a threefold per-employee gap between disciplined and typical buyers, the explanation is almost entirely plan-tier opacity, and negotiation — or the absence of it.

The companies paying at the top of the range aren't buying a different product, and they aren't simply bigger. They're buying the same tool, at the same scale, negotiated differently — or never negotiated at all. The difference comes down to when they last reviewed the contract, how many seats they provisioned versus actually use, and whether anyone in finance ever questioned it. They have no way of knowing they're overpaying, because vendors publish a generic number if that. They never publish what cohorts pay.

Note on sample: We deliberately excluded consumption-priced cloud vendors (AWS, OVH, Scaleway, Gandi) from this figure. Their bills scale with usage, not seats, so a per-employee benchmark doesn't isolate pricing discipline the way it does for SaaS. Measured on the same basis, they'd add a further €7.1M — but we're reporting that separately rather than folding it into the headline number.

The €67M overpayment pool isn't shrinking as the market matures — it's stubbornly flat year-on-year, even as median SaaS rates fell. Composition rotates between vendors (HubSpot's gap widened +€1.1M, Atlassian's narrowed −€1.4M), but the total dispersion holds. Overpayment is structural, not a transient inefficiency.

By market: The price-discipline gap travels across borders

Wherever you look, the same tool gets bought at wildly different prices, and no market has solved it. Figma is a clean test case — one product, a complex pricing plan modelled around different types of seats and contract lengths, sold to design teams everywhere. Yet in every focus market the median company pays ~200/month while the top-paying quartile pays ~€1,500 — a 6–9× spread inside every single geography, for a single vendor. The UK is the most disciplined: highest median but the narrowest spread and fewest extreme overpayers. France and Germany run lower medians but a longer tail of companies overpaying badly.

Overall, the average overspend combined with market-level data points towards €1.6 billion of overspend in France.

2. Duplicate tools: Behind the scenes spend, in every team

Key takeaway: Tool duplication costs companies European companies €3 billion a year

Spendesk analysed subscription overlap across software categories in its transaction data. The analysis found roughly €15 million a year in recoverable spend from duplicate tools across the sample, extrapolating to €3 billion at a European level. More than two-thirds of that waste concentrated in AI and design software — categories where 72% and 64% of buyers, respectively, pay for two or more competing products at once.

72% of companies using AI tools pay for two or more competing products. In design tools, it's 64%. When companies buy software, they rarely rationalise. What they do is add. Duplication isn't the mark of a few poorly-governed outliers; it's the statistical norm across almost every category we looked at.

The categories with the highest duplication rates are also the fastest-growing ones: AI and design. That's not a coincidence. Fast-moving categories attract trial-first buying behaviour, and trials have a habit of becoming permanent subscriptions nobody gets around to cancelling.

For finance, the takeaway isn't "stop bad actors." It's "fix the process." Without structured vendor rationalisation, every team's default behaviour is to keep adding tools and when the time comes to reduce budget, it’s easier when you know where to start. 

We're not immune to this ourselves, at Spendesk. We pay for both Dust and Claude, two AI assistant tools, because we use them for different jobs. The difference is that finance knows about it and signed off on it; nobody's quietly expensing duplicate licences. We have guardrails for exactly this reason and if we ever need to drop our costs, we know where to start. 

Year on year: Duplication doubled, as people commit more and more to experimenting with AI, but haven't decided which tool to commit to

A year ago the typical LLM-using company ran a single model — only 29% ran competing ones. Today the median is two, and 65% run 2+ tools. Multi-model adoption more than doubled in twelve months, going from the exception to the norm — and the duplicate spend behind it grew from €0.3M to €4.5M as pilots turned into paid seats. This tracks with models specialising by discipline: a company might standardise on one for general work, another for coding, a third for cross-team reasoning.

By market: A story of scale versus discipline

France holds the largest euro pool yet is the most disciplined. France has the lowest duplication rate of the four — 63% of software spenders run 2+ competing vendors in a category, against 74–77% in Germany and the UK. Germany and the UK are the real duplication hotspots, and spend concentrates hardest in the UK: every duplicating UK company leaves ~€12,400 a year on the table, versus ~€9,800 in Germany and ~€9,100 in France.

Taken as a general average to avoid skews for the smallest companies and extrapolating to the number of SMEs in France, the latter points to €272 million left on the table, whilst for the UK, the duplication bill would reach €496 million.

3. European AI consumption: Huge increase, huge waste

Key takeaway: European AI bills can vary by more than 7x month to month, making them difficult to forecast

AI spending is highly volatile across European companies. Spendesk analysis shows the median company’s monthly AI bill varied by approximately 62%, while its highest monthly bill was more than seven times its lowest. Among the 1,347 companies with at least three consecutive months of AI spend, 60% showed high month-to-month variance, creating a forecasting problem for finance teams.

Better AI spend tracking is a savings lever in its own right: rogue token usage only becomes visible and controllable once someone is actually watching the variance.

The median company's monthly AI bill swings ±61% month to month, and 60% of companies with otherwise steady AI usage carry high-variance bills. Usage-based pricing is a feature for vendors. For finance teams, it's a forecasting problem.

Out of the 2,500 company sample, we analysed 1,347 companies which had at least three consecutive months of AI spend and measured how much their bills moved. For the typical company, the highest month is more than 6.5× the lowest — a swing no fixed budget line anticipates. This kind of volatility won't show up in a traditional SaaS budget model, because those models assume fixed monthly costs. AI doesn't behave like a subscription. It behaves like a metered utility with no meter on the wall.

What to do about it: the only reliable response to usage-based billing variance is a spending limit with an alert layer on a smart card, not a budget line. That's where tracking pays for itself — a team that can see the variance in real time catches the rogue token run in week one, not at month-end close. Get that in place, and you liberate the AI line from guesswork, instead of discovering the swing after it's already happened. Finance teams that treat AI like a regular subscription are systematically under-controlling their fastest-growing, least predictable line of software spend.

By market: AI bill volatility does not care about geography

Where consolidation and price discipline varied sharply between markets, bill unpredictability does not. 

Every market sits inside a narrow 55–65% band of median variation, with 55–62% of companies carrying high-variance bills. There is no disciplined market to point to, and no outlier to explain away. 

The small differences run in a familiar direction: 65% median swing in the smaller "rest of Europe" markets — the most volatile group. Germany is the calmest (56%) and the UK the most exposed (highest high-variance share at 62%), but "calmest" still means more than half of companies have AI bills that swing by over half month to month. 

The flatness is the point: consolidation and pricing gaps are behavioural and closable with discipline; volatility is structural — a property of usage-based billing itself. That is why it shows up near-identically in Paris, London, Berlin and everywhere between, and why you cannot budget your way out of a metered bill — you can only put a control layer on it.

A spending limit with real-time alerts works in every market because the problem it solves exists in every market. Spendesk helps mid-size European companies solve their spend management issues by bringing all of their spend in one place, so finance teams gain control and visibility without becoming a bottleneck.

4. How Spendesk can help control spend

The answer is not to stop teams from using software or AI. It is to make that spend visible, intentional and tied to business outcomes. That starts with one clear view of subscriptions, invoices and card payments, with every tool linked to an owner, a team, a purpose, a renewal date and, where possible, actual usage. Finance can then see where duplication, unmanaged renewals and unexpected AI usage are building up before they become part of the run rate.

Control matters most when spend is dynamic. For fixed SaaS, approvals and renewal reviews can prevent unused seats and duplicate tools from rolling on. For usage-based AI, spending limits and real-time alerts can flag unusual increases early. This gives finance a practical control layer without turning every purchase into a bottleneck.

But visibility is only useful if it leads to better decisions. The next step is to connect software and AI spend to business outcomes: the time saved, output generated, customers supported or revenue enabled. A tool may be worth keeping even if it overlaps with another, provided its role is clear and its value can be measured. The question is not simply “what can we cut?” but “what is this spend helping the business achieve?” That is where Spendesk fits, bringing payments, approvals and reporting together so finance and teams can make decisions from the same view of the facts.

With the right controls in place, finance can support experimentation without allowing it to become invisible spend. Teams keep the tools they need, while finance keeps the context, guardrails and confidence to scale what works.

Get in touch with the Spendesk team to learn more.

Frequently asked questions

What is SaaS spend management, and why does it matter for finance teams?

SaaS spend management is the process of tracking, approving, controlling and reviewing software subscriptions across a business. It gives finance teams visibility into who owns each tool, what it costs, whether it is being used and when it is due for renewal. This helps businesses identify waste, benchmark pricing and prevent unnecessary renewals.

How much do European companies spend on SaaS software?

Spendesk research reveals the average European company in the sample spends almost €250,000 a year on software. 

How can finance teams track SaaS subscriptions across multiple departments?

Finance teams should create a central view of invoices, card payments and recurring subscriptions, with each tool linked to a department, cost owner, usage information and renewal date. This replaces fragmented spreadsheets with a consistent process for monitoring software spend, preparing for renewals and identifying opportunities to consolidate or renegotiate.

How can companies identify and eliminate duplicate or zombie subscriptions?

Companies can compare vendors by category, review usage and ownership, and identify subscriptions with no active users or clear business purpose. In the sample, 72% of companies using AI tools paid for two or more competing products, while 64% of design-tool users had parallel subscriptions. Approval workflows, named cost owners and pre-renewal reviews can prevent unused or duplicate tools from quietly continuing to charge the business.

How can virtual cards and spend controls help manage SaaS subscriptions and AI costs?

Virtual cards and smart-card controls can make recurring and usage-based software charges easier to identify and manage. Finance teams can set spending limits, monitor transactions in real time and receive alerts when costs rise unexpectedly. For fixed SaaS subscriptions, renewal reviews and clear ownership help prevent unapproved charges from continuing.

Why is AI spending harder to forecast than traditional SaaS spending?

Traditional SaaS is usually billed at a relatively fixed monthly rate, while AI tools often charge according to usage. In the sample, total AI spend grew by 340% in twelve months, while the number of companies buying AI barely changed. AI bills can also vary substantially from month to month, with the highest monthly bill reaching up to eight times the lowest, making them difficult to manage with a traditional fixed budget.

How can finance teams control variable AI spending and usage-based bills?

AI spend should be managed with usage visibility, spending limits and real-time alerts rather than through a fixed budget line alone. These controls help finance teams identify unusual token usage or automated activity early, before a high-cost month reaches the end-of-month close. A smart card with spending limits and alerts can provide an additional control layer.

What should finance teams look for in SaaS spend management software?

The most useful capabilities include centralised subscription visibility, cost-owner and department allocation, approval workflows, renewal alerts, recurring-payment controls, usage monitoring, spending limits, real-time alerts and reporting. For AI tools in particular, the system should provide consumption intelligence, helping finance teams move from simply recording software spend to actively controlling usage-based costs.

Sample & method

2,500 active European companies, June 2025–June 2026. Nine industry verticals. Four size bands. The composition mirrors Eurostat's SME distribution: 91% of the companies in our sample have fewer than 250 employees — because that's the reality of the European business landscape.

To model the directional, market level numbers, we used Eurostat Structural Business Statistics, EU-27 enterprises with 50-249 persons employed (2024) provisional data which showed 250,000 EU companies with 50 to 249 employees. 

To model France's directional numbers, we employed Eurostat's most recent data (2022),pointing towards an approximate number of 30,000 businesses with 50-249 persons employed. 

For the UK's directional numbers, we employed the House of Commons Library's most recent data (2025), pointing towards approximately 40,000 businesses with 50 - 249 persons employed. All numbers were rounded up to the closest round figure.

Method

To measure how European companies buy and spend on software, we built the vendor universe bottom-up from actual payments rather than starting from a pre-defined list of "known" SaaS brands. Working from a fixed panel of 2,568 European companies (14 countries, active on both card and invoice rails over June 2025–June 2026), we took every supplier each company paid across the twelve-month window and let the data reveal the vendor set.

Two payment rails carry different information, and we used both:

• Invoice / wire payables carry a normalised supplier identity, which we resolve to a single canonical vendor (collapsing regional legal entities and duplicate records — e.g. OpenAI's four internal supplier IDs merge into one).

• Card transactions carry only a free-text merchant descriptor (e.g. OPENAI *CHATGPT SUBSCR, CLAUDE.AI SUBSCRIPTION), which we normalise to the same canonical vendor names.

We then removed non-software categories — advertising, HR/payroll/benefits, travel and mobility, utilities, telecom, postal, banking/insurance/leasing, hardware and retail — and kept only vendors reaching ≥ 20 distinct companies in the panel, a threshold that filters out one-off long-tail counterparties (freelancers, landlords, local services) while preserving genuine vertical and regional tools.

The result is a discovered roster of 380 software vendors, unioned across both rails: 112 vendors that appear on both card and invoice, 265 invoice-only, and 3 card-only. Blending the rails matters most for AI, design and developer tools, which are card-first and would therefore be materially under-counted in invoice-only views — for example, cards account for roughly two-thirds of Anthropic's spend and the large majority of GitHub's.

Caveats. Vendor classification is rule-based; we estimate a residual ~10–15% of the vendor tail is non-software that evaded the category filters (IT-services resellers, EOR providers, wholesale electronics). Card attribution depends on merchant-text matching, so a small share of card SaaS with cryptic descriptors is undercounted (never mis-attributed). These do not materially affect the per-vendor and AI-specific findings that follow.