AI token costs are becoming a CFO problem, Accenture says

AI token costs are turning into a budget problem for companies trying to expand artificial intelligence, according to Accenture. The consulting firm says many finance chiefs are finding that AI pilots can become expensive once usage spreads across employees, products and autonomous agents.

In a report titled “AI is on your P&L. Most companies are only reading half of it”, Accenture Chief AI and Data Officer Lan Guan said companies often underestimate the financial demands of AI at scale. Guan told Fortune that clients are ready to broaden AI use, then run into an unexpected “cost wall” as token consumption drives cloud spending higher.

Tokens are the units processed when an AI system reads a prompt, generates an answer or carries out an agent task. Accenture describes “tokenomics” as the practice of connecting that consumption to business value, so companies can see which uses of AI justify their cost.

What are AI token costs?

AI token costs are the charges tied to the volume of text or data units processed by AI models. Every prompt, response and agent interaction uses tokens, so expenses can rise quickly when a tool moves from a small test to broad use across a company.

Guan said many large-company CFOs see only the final bill and lack a clear view of which teams, products or AI agents are using the most tokens. She told Fortune that some finance leaders say they are “walking in the dark” because they cannot trace the spending to specific activity.

Accenture cited a retail client that used an AI recommendation engine in a small number of pilot stores. According to Guan, sales rose, but the client also faced a monthly cloud bill in the millions, with token use as a main driver and a cost structure the company had not planned for.

The issue comes as corporate AI spending continues to rise. Guan cited Goldman Sachs as saying AI-related spending is on pace to exceed $800 billion in 2026, while Accenture said only 23% of C-suite leaders it surveyed reported broad and sustained business value from AI across their organizations.

Accenture has also applied the cost-control work inside its own operations. The firm said one internal platform handles about 8.7 trillion tokens each week, and it built an “AI Token Navigator” to send work to different models based on how complex and important the task is.

Guan said that routing can cut costs to about one-sixth of the expense of using only top-tier models, which are typically the most expensive. Accenture estimates that just 10% to 20% of enterprise tasks need frontier or near-frontier AI capability.

How does Accenture say CFOs should control AI token spending?

Accenture is advising companies to manage tokens as an enterprise discipline rather than a technical side issue. Guan described a three-part approach for CFOs: see usage, treat inefficient architecture and manage behavior over time.

• Accenture says companies first need visibility into where token use occurs, instead of relying on aggregate cloud bills.
• The firm says companies should route work to the right model and identify waste in AI system design.
• Accenture says finance, technology and cybersecurity teams need to coordinate, because added controls can also affect AI costs.

In one internal analysis, Accenture found use of a single AI tool rose 113-fold over 10 weeks, with 19% of users responsible for about 80% of spending. Guan said cheaper and easier AI access can lead employees to use the most powerful models more often, increasing consumption even when individual tasks become less expensive.

Guan told Fortune that AI expansion does not have to produce unchecked costs, but companies need stronger controls around usage. For CFOs, Accenture’s message is that AI spending now belongs on the profit-and-loss statement, with token use treated as a measurable operating cost.

This story draws on original reporting from Fortune.