%201.webp)
Only 3% of FTSE 100 companies reported a specific financial return or outcome from artificial intelligence in their 2025 annual reports, according to Flexera. Its analysis also found that just 2% disclosed a specific AI spending figure.
The findings highlight a gap between the number of large listed companies that say they are using AI and the number giving investors clear financial evidence of what those efforts cost or deliver.
Flexera's review of FTSE 100 annual reports found that 36% reported substantive AI deployment or use. Some 32% linked AI to a specific business outcome, but none identified AI investment as a separate budget or cost. More than half the index, 54%, identified AI as a material or company-specific risk in their reporting.
Disclosure gap
The data comes as companies face closer scrutiny over the economics of AI projects, particularly as software suppliers introduce charging models more closely tied to usage. The issue is becoming more acute as businesses deploy AI agents that can trigger multiple billable actions while carrying out a single instruction.
The review measured public disclosure rather than a company's internal ability to calculate returns from AI. Researchers recorded only figures explicitly stated in annual reports and did not estimate spending or infer outcomes.
Marlon Oliver, Senior Vice President, EMEA, at Flexera, said: "Companies are burning through annual AI budgets in months, and always-on agents will put those budgets under even more pressure. Having visibility of that spend and controlling it is essential, but it is only the starting point.
"The next question is what the business is getting in return. Token consumption, compute and licence numbers tell you what AI costs and how much it is being used. They do not tell you whether it is creating value.
"Accountability matters as well. Organisations should know whether the AI licences and agents they are paying for are actually being used, but utilisation alone is not success. A stronger test is whether the business owner receiving the benefit would be prepared to absorb that AI run-rate into their own budget.
"Ultimately, AI investment needs to connect to business KPIs, whether that is downtime reduced, tickets resolved, claims processed or revenue generated. When shareholders are told AI is improving productivity, they should be able to see what that improvement is worth, what it cost and whether the economics are improving."
Cost tracking
The spread of AI across software licences, models, data and computing resources makes it harder for IT teams to isolate spending. Consumption-based charging can also make costs less predictable than conventional software subscriptions.
That challenge becomes sharper when autonomous agents are involved. A single request may lead an agent to make repeated attempts to complete a task, with each attempt drawing on additional computing resources and increasing the bill.
Steve Daheb, Head of AI GTM Strategy at Flexera, said: "AI has become part of the core enterprise technology estate, but its economics are much harder to see. Costs span applications, agents, models, data and compute, accumulating through licences and consumption-based services. One company I recently spoke to discovered after the fact that just four users were responsible for 35% of its AI costs.
"Agents make those economics more complex because they can initiate their own workflows and consumption. A single instruction can set off repeated attempts to complete the work, each drawing on more computing power and adding to the bill. It's an AI cost domino effect.
"The next challenge is accountability and ensuring AI doesn't become an enterprise cost black box. Organisations need to connect what they are consuming and spending with the outcomes AI produces. That understanding should give businesses the confidence to invest, knowing which uses are creating value and where costs need attention.
"Companies are finding their footing when it comes to disclosing AI risk. As machine-driven consumption scales, they need to bring the same rigour to understanding and disclosing AI costs and value. If they want shareholders to keep backing their AI ambitions, they need to put financial evidence behind their progress."
Reporting limits
The methodology covered FTSE 100 companies as of early September and counted a company as reporting on AI only when its annual report included an explicit reference to artificial intelligence or a clearly identifiable AI technology. Search terms included artificial intelligence, generative AI, large language models, machine learning, AI assistants, AI agents, and named AI platforms or services.
Because the categories overlap, the figures do not imply a single uniform approach to disclosure across the index. Some companies described operational use of AI, others referred to products or services that use the technology, and a smaller group attempted to tie AI to measurable business outcomes.
Direct reporting on financial return and expenditure remained largely absent. For investors, that leaves a limited basis for comparing whether AI investment is producing measurable gains across some of the UK's biggest listed companies.
For companies, it also raises a practical management question for finance and technology teams: whether they can trace AI costs back to specific users, tools or workflows before those charges appear in annual reporting. As Daheb put it, "One company I recently spoke to discovered after the fact that just four users were responsible for 35% of its AI costs."