From tokenmaxxing to valuemaxxing

In the last few months, several leading tech firms have made a decisive move away from 'tokenmaxxing' – a trend whereby enterprises measure AI usage based on the volume of tokens consumed by employees.

In May, Amazon quietly retired an internal leaderboard that ranked staff by how many AI tokens they used, while Accenture capped routine use of AI tokens in June after one internal tool saw usage rocket 113-fold in just ten weeks. More recently, Microsoft updated internal guidance this August specifically to curb AI token spend.

Speaking on the ITPro.Podcast, Ninox CEO Frank Böhmer said the wider industry’s gamification had pushed employees to chase visible token counts out of stress and pressure rather than genuine performance.

While his company had avoided this practice, this well-intentioned but flawed metric has backfired elsewhere, incentivizing staff to optimize for volume over efficient, well-scoped use, leading to costly token burn on low-value tasks.

A model lacking economic sustainability

Rewarding staff for how much they spent on tokens was never going to be economically sustainable, says Stewart Buchanan, research VP in Gartner’s CIO team. The real-world consequences of unmanaged consumption are stark: Uber exhausted its entire 2026 AI budget by mid-April, while Dell saw a single 'super user' developer run up a bill of $3,400 in just 24 hours due to high token usage by the AI agents they were running.

This financial volatility is compounded by potential shifts in pay-per-use pricing. Anthropic, for example, announced it was separating human interaction from agentic and API use in its subscriptions. While this move has currently been paused, Ashish Nadkarni, leader of IDC’s enterprise infrastructure global research domain, believes these kinds of pricing shifts will compel CIOs and tech leaders to become much smarter about how they allocate budgets to developer activities.

Valuemaxxing and the rise of tokenomics

Understandably, multiple companies are seeking a different approach, which for many appears to be 'valuemaxxing' – changing the core measurement from raw consumption to business outcomes. While directionally right, Mike Fuller, a member of the technical staff at the Tokenomics Foundation, believes valuemaxxing is as structurally thin as its predecessor.

"While the road to understanding value runs through unit economics, that’s only half of it," Fuller argues. "Valuemaxxing works on the numerator – what the output is worth. It says nothing about the denominator – what it costs you to produce the intelligence in the first place. That’s an engineering discipline, and where the leverage sits. We would say 'tokenomics' is next, because it covers both."

Answering what each unit of work costs and whether it landed successfully requires two critical joins:

• Spend to workload: This is being solved. The 1.5 release of FOCUS (the FinOps Open Cost and Usage Specification), expected in December and backed by the FinOps Foundation, will add native AI token tracking and a price sheet dataset. This will allow businesses to compare token spend across different AI providers in a consistent format.
• Workload to outcome, which remains a blind spot. "The industry is still guessing here," Fuller admits. "There’s no standard for connecting a workload to a business result, and there may never be a universal one, because the result differs by business."

To establish this second join, organizations must tie every prompt to a business outcome, tracking cost and revenue per customer, transaction, or item. "‘Miles per gallon’ is often used as a proxy for this," notes William Fellows, research director at 451 Research. "But the problem when it comes to tokens is there’s no industry agreement yet on what constitutes a mile, a gallon, or indeed the fuel itself."

Shifting employee mindsets

While the industry establishes these frameworks, organizations can begin shifting staff mindsets away from high-volume token consumption immediately, and education is the primary lever for behavioral change, says Fuller.

When staff understand that the goal isn’t outcomes at any cost – and are equipped to assess the choices that change what an outcome costs – behavior changes naturally. Organizations must transition from measuring sheer usage to answering harder questions, Fuller says:

• Where does AI add genuine value?
• What investments make financial sense?
• What happens to the staff time AI frees up, and who’s responsible for redeploying it?

Buchanan adds that tech leaders must train staff to identify the most valuable use cases while actively discouraging uneconomic AI habits. “For instance, a simple, rule-based engine can consistently deliver deterministic outcomes without burning costly tokens on complex reasoning and inference.

“Staff must also learn to discourage perfectionism; developers and analysts frequently run the same prompt repeatedly to perfect an answer, when they should learn to stop at the first adequate response that can be refined manually at a lower cost.”

Redefining the operating model

To manage this spend sustainably, enterprises must address where AI budgets actually sit. For HPE CEO Antonio Neri, AI agents should be categorized alongside human resources rather than traditional IT infrastructure.

"I don’t think of AI agents as an IT cost," Neri told ITPro. "I think about the cost of the workforce because, to me, an agent is no different than any other employee I have to hire… it's going to cost me a number of tokens to train an agent to drive the best productivity. If I'm going to spend a million dollars to train an agent, it has to be way more productive than a human. Otherwise, why am I doing that?"

However, Buchanan warns against over-simplifying this comparison. "We personify and anthropomorphize AI at humanity’s peril," he cautions. "People and AI aren’t identical and interchangeable – people think on a few thousand calories a day, while AI data centers consume gigawatts. Neither HR nor IT manages the business nor its budgets, so we need deeper integration with business financial planning and analytics."

AI spend must reach board-level discussion as a capital allocation question, Fuller argues. Boards must ask what they’re committing to, over what term, on what pricing assumptions, and what their financial exposure is if a single provider changes its terms.

The path forward for CIOs

While "cost per outcome" is the ideal destination, most CIOs won’t realistically achieve this level of granular tracking within the next 12 months. Instead, the immediate, pragmatic goal for tech leaders must be twofold: identify which AI use cases are actively succeeding against valuable business outcomes, and clearly name the pilot investments that have yet to prove their value.

Buchanan’s advice is to skip the magic formula; CIOs must partner directly with business leaders to control costs in relation to strategic value. Failing to do so risks creating severe corporate governance challenges through shadow AI across the business.

By focusing on unit economics, shifting employee behaviors, and integrating AI into the broader financial planning model, enterprises can move past the chaotic era of tokenmaxxing and build a sustainable, value-driven AI strategy.