How much are you spending on AI?

As AI adoption accelerates, Quentin Vigneau, CPO at Spendesk, explores why finance teams need greater visibility into AI spend and how connected spend data can support better investment decisions

AI tools now account for 38 per cent of all technology spend processed through Spendesk, up from less than one per cent four years ago – Quentin Vigneau, CPO, Spendesk

European businesses almost doubled their spend on AI tools in 2025, according to the all-in-one spend management platform Spendesk. 

Perhaps that doesn’t surprise you, given the rapid rate of adoption, particularly of AI assistants and AI agents.  But this figure might: 70 per cent of those AI purchases were subscriptions and the problem with subscriptions is that they frequently obscure the real cost of compute… until multi-step workflows crash into your spend cap, trigger overuse messages or force tier upgrades.

It’s probably why a survey by FinOps Foundation found that AI costs overran enterprises’ quarterly budget projections in 73 per cent of cases between 2025 and 2026. It’s said that Uber bust its annual AI budget in four months. 

So, ‘how much are we spending on AI tools, and who owns them?’ is becoming a pretty pertinent question for finance teams, given that AI was supposed to reduce opex, not increase it. And it’s the one that Spendesk is helping clients answer right now with Spendesk AI Connect. 

Released in July, Spendesk AI Connect does what it says on the tin: it’s the infrastructure layer that connects finance teams and business data with AI assistants and AI agents that are already being used by the organisation. Among other things, it will allow teams to work out just how much those tools are costing, by querying the spend data they already hold in Spendesk. 

“By allowing AI assistants to query selected, permissioned Spendesk’s live data, companies can analyse AI-related spend, identify recurring payments and review relevant card owners, expiry dates and recent activity,” explains Quentin Vigneau, who came on board as Chief Product Officer (CPO) alongside Alan Wright as Chief Technical Officer (CTO) earlier this year. 

“Instead of discovering overspend during quarterly reviews, your finance teams can monitor continuously and tackle it early. And as the market evolves, we're developing even more granular analytics to help companies understand not just what they’re spending, but how that spend is distributed across suppliers, teams and recurring payments..”

When it comes to AI agents and assistants, that’s critical because while we’re finding more complex tasks for these tools to do, they’re not necessarily the ones with the most valuable output for the business. 

The company’s own research indicates that by far and away the most popular of those is Anthropic’s Claude, followed by ChatGPT, Gemini and Microsoft Copilot. But it revealed there’s no one killer app for finance teams – and that was the starting point for Spendesk AI Connect.

Built on open standard Model Context Protocol (MCP), it allows spend data to be discovered and read by any AI assistant, agent or AI orchestration platform.

“The shift we're seeing in customer behaviour reinforces why this matters,” says Vigneau. “AI tools now account for 38 per cent of all technology spend processed through Spendesk, up from less than one per cent four years ago. As AI becomes embedded across every part of the business, finance teams are increasingly working across multiple tools and assistants rather than a single platform. 

“Different tools will be better suited to different tasks, and finance teams are already experimenting with a range of platforms. That's why supporting third-party assistants like Claude, Dust and others is so important. Our customers are already using these tools, and they want them connected to reliable, up-to-date financial data,” says Vigneau.

“By adopting MCP, we're following an emerging industry standard that makes those integrations easier and more secure.”

The Spendesk 2025 Top CFO Tools Report showed that finance departments are being pragmatic and reassuringly cautious in deciding what to delegate to AI assistants, playing to their speed and processing power rather than their intelligence.

“Finance teams are comfortable using AI to retrieve information, identify patterns and accelerate analysis but financial decisions still require accountability,” says Vigneau.

Specific, well-defined tasks that save teams time and reduce manual work, such as preparing financial presentations and meeting notes, helping compile reports and data analysis, industry research and document retrieval, do not represent a revolution in business processes. There’s nothing groundbreaking to see here. But it does dramatically improve operational intelligence and a CFO’s response time.  “Instead of waiting for the next reporting cycle, CFOs can get answers immediately and focus on decision-making, planning, and supporting the wider business,” says Vigneau.

“The questions we're seeing first [through Spendesk Connect AI] are very practical ones. What's our current cash position across entities? Which invoices are at risk of being paid late? How does departmental spe across periods?  These aren't new questions but until now they often required someone in finance to pull together the data manually. Spendesk AI Connect makes that information instantly accessible.”

The cautious ‘test-and-learn’ approach being adopted by most teams and the comparatively limited number of tasks they are willing to trust to AI is also reflected in the approach taken to developing Spendesk AI Connect.  

“The opportunity with MCP is that it keeps the human in the loop and only access the data its user has permission to access. This we call “read only” capabilities. We’re also working on enabling write-capabilities later this year,” says Vigneau. “The system will always operate on the principle that the assistant proposes and a human approves, before applying the appropriate permissions and confirmation flow.”

As a regulated financial platform, Spendesk is designed to meet stringent compliance and security standards including ISO 27001:2022 certification, GDPR compliance, MFA, PSD2, PCI-DSS and end-to-end encryption. Security is central, says Vigneau.

“Spendesk AI Connect uses an OAuth-based permission model [the industry standard protocol for authorisation], ensuring any AI assistant can only access data and perform actions that are an explicit subset of the authenticated user's own permissions. Rather than acting as an independent system, large language model terminals serve as an extension of the user's existing capabilities within Spendesk. 

“This approach reflects how finance organisations already operate. CFOs and controllers want AI to eliminate friction and surface better insights, while preserving the governance and oversight that underpin financial control. 

“By embedding authorisation and confirmation into the product experience, we can give customers confidence that AI is accelerating finance without bypassing the controls that matter. That's ultimately what builds trust and encourages adoption.”