As companies increasingly adopt artificial intelligence (AI), many Chief Financial Officers (CFOs) are facing an unexpected challenge: accurately assessing their organizations' AI expenditures. An unpublished study by a consulting firm reveals that only 26% of companies fully understand their spending on AI technologies.
Half of the organizations have only partial insights into their costs, while 22% learn about their expenses only after receiving bills or lack sufficient transparency altogether. Steve Chase, the global lead for AI at the consulting firm, notes that the issue stems from the emergence of a fundamentally new type of resource whose consumption is growing exponentially.
The root of the problem lies in the changing pricing model. Unlike traditional software where corporate clients would typically pay a fixed fee, many AI vendors now charge based on actual usage. This new model measures costs through tokens—units of text processed by AI models when generating responses or performing tasks.
Prominent AI model developers, including OpenAI and Anthropic, along with major corporate software providers like Microsoft and Salesforce, have adopted this approach. While it allows vendors to better align revenue with computational costs, the financial risks have largely shifted to clients.
Chase highlights that some companies have exhausted their annual budgets for tokens and cloud computing within just a few months. In one instance, a KPMG client saw its token consumption surge sixfold.
Life360, a company offering geolocation and digital security services, is also grappling with this issue. CFO Russell Burke mentioned that the company is implementing tools to reduce token consumption and is reworking its AI agents' architecture, but lacks real-time expense monitoring.
The increase in costs is particularly evident in software development, where the use of AI agents for coding has led to immediate spikes in token consumption. Affirm's CFO Rob O'Hare reported that the company closely tracks AI usage almost in real-time and provides weekly expense reports to management. Despite rising costs, Affirm believes that enhanced developer productivity justifies the investment.
Consumer goods manufacturer Reckitt is similarly scrutinizing its AI usage. After launching 12 solutions for its marketing divisions, management found that some employees reverted to traditional methods, leading to reduced engagement with certain tools. Additionally, one system encountered data quality issues, prompting the company to slow its rollout.
Reckitt's CFO Shannon Eisenhardt acknowledged that the payback periods for some projects have lengthened, yet this has not diminished the company's confidence in pursuing further AI investments.
Analysts draw parallels to the cloud technology boom during the COVID-19 pandemic, when companies rushed to acquire remote work software and later had to cut costs and reassess their implementations. Experts believe the AI market may undergo a similar optimization phase.
Corning, a manufacturer of glass and fiber optic products, has already limited the range of AI tools accessible to employees, concentrating its budget on a few major projects while still encouraging staff experimentation with new technologies.
In contrast, sports group Amer Sports, which owns brands like Arc'teryx and Salomon, is taking a more cautious approach. The company is exploring AI applications for processing invoices and financial reporting while deliberately slowing down implementation to prevent uncontrolled token and AI agent spending.
As AI becomes integrated into everyday business processes, companies face new management challenges: not only must they assess AI effectiveness, but they also need to monitor its consumption. For many CFOs, tracking token usage may soon be as crucial as managing expenses for cloud services and software once was. This evolving landscape suggests a need for greater financial oversight and strategic planning within the AI sector, impacting market dynamics and competitive strategies.
Informational material. 18+.
