Tokenmaxxing began as a metric to encourage AI productivity, but it became a window into artificial intelligence’s slippery costs.

So-called “tokenmaxxing” trended this year as tech companies used leaderboards to push employees to adopt AI. Productivity was measured in tokens, bits of data — words or partial words — that flow into and out of AI.

Some employees were good at it. Too good. Whether they were coders, executive teams, marketers, engineers or knowledge workers, many became masters at increasing their token usage, whether on big projects or proofing emails.

But the experiment backfired. Embracing AI at that level led to skyrocketing costs, blowing several companies’ AI budgets in a couple of months. And token use offered no visibility into what the company got out of it.

As CFOs look ahead to the next budget season, predicting AI usage is emerging as one of their most difficult challenges, alongside determining the return on investment.

It raises an important question: We’ve reached a point where you can’t avoid AI, but can you put a lid on it?

How tokenmaxxing exposed a bigger problem with AI measurement

Tokenmaxxing wasn’t a productivity fad but a sign of a structural accounting failure as companies struggled to measure value and ROI from AI usage, said John Rowell, co-founder and CEO at Revenium, a control agent that monitors AI usage and spend.

“The entire industry was basically admitting in public that AI is the first major line item in enterprise history with no unit economics,” he said.

Tokenmaxxxing wasn’t so much the disease as the fever, Rowell said. The pendulum is swinging back in reaction, with companies trying to rein in usage or cost without losing value. Some companies are rushing to cancel subscriptions or switch to cheaper AI providers, or use other tactics like shrinking context windows (which limits how much history AI reviews to calculate a response, making it less expensive).

Rowell warns that blindly yanking spending on token use kills your best work. Some of his most expensive token projects turn out to be their most valuable because they’re deep, multi-step work. In other cases, such as marketing, AI has allowed him to do that work in-house at a lower cost.

Questions to ask yourself:

• What am I actually getting out of what I’m spending on AI?
• Who are my most and least productive users? Why is there a gap?
• For my least productive users, how do I get them up to speed with my most productive users? What are they working on that’s not producing the same level of results?

AI usage needs the same oversight as any other workforce

Finding the balance is a legitimate problem as companies transition to a world of AI. One strategy is to view AI as a workforce and manage it like one. That especially goes for agentic AI that can independently perform multi-step tasks.

“You’ve essentially got, now, employees with an unlimited credit card,” Rowell said, pointing out their ability to rack up costs through tokens. “Without monitoring, you don’t know the results of what they’re getting, what they cost you, what the outcome is or the value of what they’re producing. There’s no expense report, there’s no approval process. It’s the wild, wild west.”

Spending can get out of control quickly. For instance, Rowell said one of his clients was gone for the weekend and came back to find they had spent $47,000 because two AI agents got into a fight, running up the token bill. ​

Instances like that are not technical failures but management ones, he said. Having a system to monitor what your company is getting out of AI is key to measuring its value, ROI and keeping costs in check.

Questions to ask:

• What safeguards do you have to monitor AI usage?
• How are you tracking your AI spend?

CFOs need better ways to connect AI spend to business value

With the continued rapid advancement of AI, the CFO's task of demonstrating its value will evolve.

Showing AI gross margin by product will stop being an exotic bit of information and become more of a standing one boards expect as adaptation deepens, said Rowell. (See Pt. I and Pt. II of our AI Risk guides for more questions boards will want answered.)  

The key is to focus on results, not tokenomics or the AI economy. Stop counting tokens and start asking what the puts and takes are for the customer, whether internal or external.

While this is an emerging technology, the underlying game is accounting, focused on margins and ROI. AI’s success hinges on whether it improves the numbers.

“No one goes, ‘Hey yeah, let’s go accounting!’ But accounting is the game right now for this,” Rowell said. The big questions include whether the company is making money from AI, whether its use makes sense, how it affects your workforce, and how you think about it in a way that gives you answers, not just guesses.

“Right now, for more people, it’s guesswork,” he said.

Questions to ask:

• Did the token spend clear your margins?
• What was the outcome, and was it satisfactory for your business?
• How does AI use affect your workforce?
• Do you have the information you need? If not, how can you get it?