As companies use artificial intelligence more than ever before, they are also tracking how employees are using AI with unprecedented detail. Yet many CEOs are hopeful, but still can’t say, if that makes employees more productive.
More than two-thirds of businesses still rely on metrics, such as time saved or cost reductions, rather than financial results to assess AI’s return on investment, according to a 2026 survey of the top 100 AI business leaders from ModelOp, an AI lifecycle management and governance platform. ModelOp refers to the gap between AI functionality and measurable return on investment as the “AI value illusion.”
“Almost every Fortune 500 is tracking the use of AI,” said Jim Olson, ModelOp’s CTO. But very few follow what the board really cares about: whether that spending provides a return on investment,” he said.
With the tools that appear Microsoftenterprise customers can track how AI tools are being used across their organizations, including active users, data volume, and agent activity over time. “Customers start with acquisition and engagement metrics and then continuously connect those insights to broader product and business outcomes,” said a Microsoft spokesperson.
Each interaction comes with a cost. Those costs are measured in “tokens,” which AI unit companies charge for each piece of text or data processed, turning all information into a trackable cost. But while companies have detailed visibility into how much AI is being used and how much it’s costing, they have far less clarity on who is using it effectively or whether it’s improving their performance.
Many organizations remain in the experimental phase instead of implementing AI on a meaningful scale. Still, most companies (64%) say AI is driving innovation, but only 39% report a measurable impact on revenue, according to McKinsey’s report.
Sameer Gupta, America’s AI finance leader at EY, said that in his experience, companies are more likely to measure the use of AI at the team or role level. “The focus is on results and efficiency, not monitoring each individual,” he said.
That often means comparing patterns across teams or roles rather than examining individual employees directly.
“The biggest challenge is not to measure the use, to show what is being done,” said Gupta. “Leaders can see where AI is being used and where productivity appears to be improving, but distinguishing AI as a key driver is difficult.”
‘Tokenmaxxing’ is a new AI line item
In some workplaces, the use of AI is starting to feel less like a tool and more like a competition to prove employee productivity, with internal systems ranking employees on leaderboards based on how much they use AI, and internal tracking revealing very large spikes in usage for individual employees. That appears to fuel what some in the industry call “tokenmaxxing,” where employees try to increase their use of AI to demonstrate productivity. But critics warn that more information does not lead to better work, raising the risk of AI becoming a proxy for work rather than results.
“The use of AI is a very poor manufacturing practice,” said Ravin Jesuthasan, senior partner and global innovation leader at Mercer.
“They see the use of tokens … but not really what those tokens are used for,” Olson said.
Esteban Sancho, CTO of North America at Globant, a digital transformation consultancy, says there’s a good reason why employees may feel pressured to collect tokens as AI becomes more widespread throughout the business. “If you’re not using tokens, you’re probably not working,” he said, referring to parts of the business where AI agents now manage core processes.
The use of AI is built into how work is delivered, priced, and analyzed. “Token costs are now a common factor in our ROI calculations,” Sancho said. Those costs are considered part of the company’s cost of labor and infrastructure. All AI work flows through an internal platform that tracks token usage, usage patterns, and costs across teams and projects.
“Project leaders have access to information that a team member is using,” Sancho said. He added that low utilization is not automatically considered a performance issue but is used to identify inefficiencies.
The use of tokens is calculated directly into the project budget and return on investment, and companies can continue to adjust models, budgets, and workflows based on where AI generates the most value. Teams can be reorganized around AI, creating what Globant calls AI pods where the technology delivers the most measurable benefits.
Those changes are now translating into revenue for Globant. AI-driven services that did not earn money last year reached an annual run rate of $20.6 million by 2025, and the company expects that to grow to $100 million, according to Sancho.
Coinbase announced on Tuesday that it is reducing its headcount by 14% and eliminating several layers of management, a restructuring that will include what its CEO Brian Armstrong called the adoption of “AI-native pods” with limited human talent controlling a fleet of AI agents. It will also include “testing” in one-person teams, he wrote in a post to employees — for example, a single role for an engineer, designer, and product manager.
It’s easier to scale AI agents than workers
What is ironic at first, in the anxious days of deploying AI to the workforce is that it is easier for companies to measure results when the work is done by AI systems than by humans.
At Salesforce, executives argue that the role of AI agents is leading the industry to move beyond tracking the use of AI to measuring whether work is actually being done. Both of these metrics matter, but they should ultimately map to a measurable ROI, such as cost savings, revenue growth, or improved customer outcomes, says Madhav Thattai, senior vice president and GM of Salesforce AI.
As the adoption of agents scales, job tracking is moving from the employee level to AI testing across the entire workflow. That measure has three layers: how much AI is being used, whether it’s ultimately completing tasks, and whether that work translates into real business results. “The power comes from connecting them, because only then do you get a complete picture of what ‘performance’ really means in the agency business,” said Thatai.
Salesforce said its platform generated 2.4 billion of these jobs, including 771,000,000 in the first quarter, up 57% quarter over quarter. In customer service, AI agents handled 129 million transactions in one quarter, and internally the company said it automated 96% of support cases and saved more than 50,000 hours of sales work.
The same changes apply to customer usage. Travel company Engine, for example, released an AI agent in 12 days that now handles 50% of chat volume while reducing hold time by 15%. At Salesforce itself, its Agenticforce system resolves 63% of customer support inquiries automatically while maintaining customer satisfaction levels comparable to human agents. Heathrow Airport saw a 30% increase in digital revenue tied to AI-driven agents, while OpenTable improved its resolution rate by 40%, according to Salesforce.
Even with these more advanced metrics, the line between tracking work and tracking employees remains blurred.
At Meta, internal systems are being tested to track mouse movements, clicks, and keystrokes to train AI systems on various sites and apps, according to an internal document viewed by CNBC. The effort is part of a broader program to train AI systems on how employees actually work, capturing everything from navigation patterns to keyboard shortcuts. The company says the data will be used to improve its models, not to assess individual performance, although the level of monitoring raises concerns about how far workplace tracking can go.
“Although many workers know this, a large minority do not and they should,” said Jesuthasan. “It is up to the organization to ensure that this is well communicated and widely understood,” he said.
