Agents Are Pushing AI Token Bills Higher: Capgemini

Nearly three in five large organisations spent significantly more on AI in the past year than they expected or budgeted. Among those affected, spending ran 24% over on average.

The findings come from The cost of intelligence: Governing AI economics in the agentic enterprise, a study by the Capgemini Research Institute.

It is based on a September 2026 survey of 500 CXO-level executives at organisations with annual revenue above US$1 billion.

About one in five organisations (21%) overshot their AI budget by more than 30%. Two thirds of those that overshot faced negative financial consequences, such as lower free cash flow or reduced discretionary spending.

Concern is rising. More than half (53%) of executives are now concerned about the overall financial cost of AI, up from 45% a year earlier. Only 56% are fully confident they can forecast how costs will change as adoption scales.

The report says six in ten organisations lack an enterprise-wide view of AI costs. Human oversight and rework are the least measured cost categories.

Only a third of executives, and one in four CFOs, are fully confident they can estimate indirect costs such as oversight, rework and governance. Half say AI spending is approved without adequate visibility of its full cost.

Agents are adding pressure. The report cites OpenRouter data showing agentic token use grew 14 times in six months, against 2.8 times for human use.

Spending Without Outcomes

Two thirds of executives say they can verify the business value of AI-assisted outcomes. Yet most organisations cannot directly match AI spending with successful business outcomes.

Some do not formally measure outcomes. Others track costs or usage but not results, or track the two separately. Even organisations with established tracking can link only 45% of AI spend to verified value.

Capgemini proposes a metric it calls AI Yield: the verified value of an AI-assisted outcome after its full safe-delivery cost. Only one in five organisations calculates it.

Demonstrating enough value to secure funding is the biggest obstacle for 57% of respondents. Employee productivity is the most cited source of value, followed by lower operating costs and faster processes. Fewer than a third (32%) cite revenue growth.

The metrics are expected to shift. Today 60% primarily measure total AI expenditure against budget, but only 44% expect to prioritise that in the next six to 12 months.

First-pass success rate is set to become the top-priority metric, followed by cost per completed task or workflow.

In response to overruns, 53% tightened spending controls or scaled back the affected initiative. Only one in five rethought tools or processes, including which activities should be AI-led rather than human-led.

More than three in five expect to adopt a multi-model strategy in the next 12 months, choosing models on cost, capability, latency and risk. Nearly three in five plan to expand their use of open-weight models.

CFOs Move In

CFOs are now more involved in approving major AI investments, the report says.

"Organisations need to stop treating its cost as a technology or procurement issue and start managing AI as a portfolio of economic outcomes," said Volker Darius, CEO of Capgemini Invent.

He said organisations need to measure cost, quality and business value together. They should also use the most appropriate model for each task and account for human oversight.

The report's recommendations include measuring the full cost and value of every AI outcome, and strengthening governance to stop AI costs from escalating.

 

Business Solution

Nearly three in five large organisations spent significantly more on AI in the past year than they expected or budgeted. Among those affected, spending ran 24% over on average.

The findings come from The cost of intelligence: Governing AI economics in the agentic enterprise, a study by the Capgemini Research Institute.

It is based on a September 2026 survey of 500 CXO-level executives at organisations with annual revenue above US$1 billion.

About one in five organisations (21%) overshot their AI budget by more than 30%. Two thirds of those that overshot faced negative financial consequences, such as lower free cash flow or reduced discretionary spending.

Concern is rising. More than half (53%) of executives are now concerned about the overall financial cost of AI, up from 45% a year earlier. Only 56% are fully confident they can forecast how costs will change as adoption scales.

The report says six in ten organisations lack an enterprise-wide view of AI costs. Human oversight and rework are the least measured cost categories.

Only a third of executives, and one in four CFOs, are fully confident they can estimate indirect costs such as oversight, rework and governance. Half say AI spending is approved without adequate visibility of its full cost.

Agents are adding pressure. The report cites OpenRouter data showing agentic token use grew 14 times in six months, against 2.8 times for human use.

Spending Without Outcomes

Two thirds of executives say they can verify the business value of AI-assisted outcomes. Yet most organisations cannot directly match AI spending with successful business outcomes.

Some do not formally measure outcomes. Others track costs or usage but not results, or track the two separately. Even organisations with established tracking can link only 45% of AI spend to verified value.

Capgemini proposes a metric it calls AI Yield: the verified value of an AI-assisted outcome after its full safe-delivery cost. Only one in five organisations calculates it.

Demonstrating enough value to secure funding is the biggest obstacle for 57% of respondents. Employee productivity is the most cited source of value, followed by lower operating costs and faster processes. Fewer than a third (32%) cite revenue growth.

The metrics are expected to shift. Today 60% primarily measure total AI expenditure against budget, but only 44% expect to prioritise that in the next six to 12 months.

First-pass success rate is set to become the top-priority metric, followed by cost per completed task or workflow.

In response to overruns, 53% tightened spending controls or scaled back the affected initiative. Only one in five rethought tools or processes, including which activities should be AI-led rather than human-led.

More than three in five expect to adopt a multi-model strategy in the next 12 months, choosing models on cost, capability, latency and risk. Nearly three in five plan to expand their use of open-weight models.

CFOs Move In

CFOs are now more involved in approving major AI investments, the report says.

"Organisations need to stop treating its cost as a technology or procurement issue and start managing AI as a portfolio of economic outcomes," said Volker Darius, CEO of Capgemini Invent.

He said organisations need to measure cost, quality and business value together. They should also use the most appropriate model for each task and account for human oversight.

The report's recommendations include measuring the full cost and value of every AI outcome, and strengthening governance to stop AI costs from escalating.

 

Business Solution