Accenture reported on September 17 that 25% of companies it surveyed had included token spending in their AI budgets, while 50% counted AI costs among their top three planned cost increases. One in three respondents had exceeded its annual token budget. Accenture surveyed executives at 17,000 companies across 17 countries, including Japan, about AI use and management.

Executives said token use had risen 78% over the previous two years, and budget estimates were often too low. Among companies facing token spending at least 25% above budget, 95% said they would increase overall AI-related spending. Accenture said this suggests companies may be able to choose how they spend on AI to reduce costs.

The survey found that only 2% of companies had established a clear link between AI use and business results. Although many said AI’s return on investment was important, only 35% said they could calculate AI costs against results. Model routing—choosing a model according to a task—was considered an important cost-control measure by executives, but only 30% said they had implemented it.

Accenture analyzed 9,368 work tasks in the public ONET database. It estimated that tasks requiring frontier models’ capabilities made up about 1% of all tasks, while 54% of AI requests were being routed to high-capability models even when the task did not require them. Chargeback, a method for allocating costs to individual users or teams based on AI use, was another possible cost-control tool, but only 7% of respondents had implemented it.

Token-management challenges were reported by 37% of companies. At 48% of companies, responsibility for managing tokens was distributed across several IT departments. Accenture said five practices could help manage token costs and business outcomes: set AI goals and accountability; assign AI costs to the relevant teams; choose models to fit each task; clarify expected value, cost and scale before adoption; and define criteria for selecting models to support effective employee use.

Accenture estimated that combining clear accountability, model routing and AI cost allocation could reduce token costs by up to 44% over the next two years while preserving business results. It said companies that clearly connect AI spending to results may have an opportunity to improve by more than 60%. The survey was conducted in July 2026 among 750 executives, including CIOs, CTOs, CFOs and CEOs, at companies with annual revenue above $1 billion. Respondents were in 17 countries, including Australia, Brazil, Canada, France, Germany, India, Japan, Mexico, Saudi Arabia, Singapore, South Africa, the United Kingdom and the United States.

Accenture reported on September 17 that 25% of companies it surveyed had included token spending in their AI budgets, while 50% counted AI costs among their top three planned cost increases. One in three respondents had exceeded its annual token budget. Accenture surveyed executives at 17,000 companies across 17 countries, including Japan, about AI use and management.

Executives said token use had risen 78% over the previous two years, and budget estimates were often too low. Among companies facing token spending at least 25% above budget, 95% said they would increase overall AI-related spending. Accenture said this suggests companies may be able to choose how they spend on AI to reduce costs.

The survey found that only 2% of companies had established a clear link between AI use and business results. Although many said AI’s return on investment was important, only 35% said they could calculate AI costs against results. Model routing—choosing a model according to a task—was considered an important cost-control measure by executives, but only 30% said they had implemented it.

Accenture analyzed 9,368 work tasks in the public ONET database. It estimated that tasks requiring frontier models’ capabilities made up about 1% of all tasks, while 54% of AI requests were being routed to high-capability models even when the task did not require them. Chargeback, a method for allocating costs to individual users or teams based on AI use, was another possible cost-control tool, but only 7% of respondents had implemented it.

Token-management challenges were reported by 37% of companies. At 48% of companies, responsibility for managing tokens was distributed across several IT departments. Accenture said five practices could help manage token costs and business outcomes: set AI goals and accountability; assign AI costs to the relevant teams; choose models to fit each task; clarify expected value, cost and scale before adoption; and define criteria for selecting models to support effective employee use.

Accenture estimated that combining clear accountability, model routing and AI cost allocation could reduce token costs by up to 44% over the next two years while preserving business results. It said companies that clearly connect AI spending to results may have an opportunity to improve by more than 60%. The survey was conducted in July 2026 among 750 executives, including CIOs, CTOs, CFOs and CEOs, at companies with annual revenue above $1 billion. Respondents were in 17 countries, including Australia, Brazil, Canada, France, Germany, India, Japan, Mexico, Saudi Arabia, Singapore, South Africa, the United Kingdom and the United States.