Since tokenmaxxing is the wrong metric for AI ROI, what is the right one?
Follow us
Add us as a preferred source on Google
The early rush to encourage employees to use AI wherever possible led to a new way of measuring productivity based on the consumption of AI tokens, or technically the underlying AI foundation model tokens. Employees who did not use AI were immediately visible through their AI consumption bill, or lack thereof, while high consumption was treated as a positive signal. The number of tokens consumed became a badge of honor.
For a while, more AI usage suggested more productivity and innovation. Some organizations even began treating AI activity as a proxy for employee engagement. This led to the phenomenon of ‘tokenmaxxing’, with the unintended consequence of AI costs growing out of control and raising a more difficult question: what does successful AI adoption actually look like and how do you measure it?
General Manager for ITSG at GoTo.
The latest Pulse of Work 2026 research points to the gap between adoption and value. While 88% of employees say AI has benefited them and estimate it saves them an average of 2.3 hours every day, four in five say they are not using it to its full potential and 69% do not fully understand how it applies to their role. Meanwhile, 60% feel pressure to use AI regardless of whether it is the right tool for the task.
Latest Videos FromTechRadar
These findings reinforce two points that should already be clear: one, AI activity is a poor proxy for business value, and two, organizations need to become more deliberate about where and how they use AI tools.
More AI does not automatically mean better work
The knee-jerk reaction to equate AI usage with success is understandable. Instant answers, automation of repetitive tasks and the removal of administrative burden from busy employees are clear and tangible benefits.
But problems emerge when AI becomes an expectation in every initiative rather than a considered decision that accounts for where human judgement is needed, which model is appropriate and what the associated cost will be.
When employers celebrate AI adoption without defining where it genuinely adds value, a problematic incentive structure develops. Employees naturally reach for AI more often, sometimes without considering whether it is the right tool, whether the output needs significant review or whether the cost of the model is proportionate to the task. Without clear parameters or support, the likelihood of misuse goes up.
Experimenting with AI should not be discouraged. In fact, it should be actively encouraged and is a crucial part of using this transformative technology for the greatest impact.
However, it’s important to also train employees so that they take a cost-conscious approach and leverage the right models for the type of task – using lower-cost models such as Gemini Flash and Claude Haiku for most tasks and only using the more expensive and powerful models such as Claude Opus and Claude
Fable for more complex tasks. The challenge comes when organizations encourage and reward use without putting the same energy into providing the proper training, guidance on cost-conscious adoption and development opportunities for their employees.
Productivity depends on workflow design
The biggest productivity gains come from identifying the scenarios where AI removes friction without compromising quality. Administrative work and removing toil remains an obvious starting point. Scheduling, email management, customer communications and routine documentation can all give employees time back for higher-order work.
From there, the real opportunity is helping employees build on their existing skills using AI to pressure-test ideas and move faster. That’s when the human-AI partnership works as it should, with technology extending what people can do rather than becoming a substitute for their judgement.
Role-specific guidance is what makes this possible. Although AI capabilities continue to improve rapidly, every function uses technology differently. A finance professional faces different governance requirements from a software engineer, and customer service teams need different safeguards from HR teams. AI training should account for such differences.
This is also why organizations should stop expecting employees to discover the best applications of AI independently. If people are spending time looking for ways to make AI useful, testing tools without guidance or working out how new capabilities fit into existing processes, some of the potential productivity gain is being lost before the technology has even delivered an output.
A better approach is to redesign workflows around genuine business outcomes. That means identifying where AI can remove unnecessary effort, then giving employees the tools and guidance to use it within the context of their role.
It also means making useful, approved AI available within the systems employees already use. When AI is integrated into daily workflows, people are more likely to apply it to safe, high-value use cases rather than seeking out unsupported alternatives.
Quality has become the missing metric
One of the highest hidden costs of widespread AI adoption is the additional burden of review and correction. AI can produce an answer quickly, but speed has limited value if employees then have to spend significant time checking whether that answer is accurate, complete or appropriate.
This is particularly important as AI-generated content becomes embedded in everyday work. A poorly reviewed output can create more work for another employee, introduce errors into a business process or undermine trust in the technology altogether. The productivity calculation therefore needs to account for what happens after the AI generates its answer.
Output quality, and ultimately, quality outcomes, should become part of the AI architecture itself. That means building review processes into workflows, introducing governance where decisions carry higher risk and ensuring employees understand the importance of validating AI responses before those responses become business decisions.
Done well, these safeguards should not stand in the way of adoption or slow employees down. They should give employees the confidence to use AI more effectively and adjust involvement where greater scrutiny or human judgement is needed. By implementing automated verification steps and role-specific parameters, IT leaders give teams the guardrails they need to drive productivity without compromising accuracy and trust.
The organizations creating value are building confidence
The biggest shift required now is cultural. Employees need clarity about when and how AI should be used, confidence that they are making appropriate decisions and practical support when new capabilities are introduced.
That support must extend beyond learning how to use a particular tool. Employees need to learn the best practices of reviewing AI-generated work for accuracy and bias, when to trust or question an output and when creativity or human judgement should take precedence.
Policies are key in achieving this by establishing consistency, but role-specific training, guidance and continuous learning ultimately determine whether those policies translate into everyday behavior.
For IT leaders, responsible adoption means creating the conditions for that confidence. They need enough visibility and oversight to manage security, compliance and quality, without introducing so much friction that employees simply look for workarounds.
The same principle applies to the technology itself. Organizations need tools that fit naturally into existing workflows and give employees the right capabilities for the work they are actually doing. AI adoption becomes much harder to manage when employees have to piece together their own collection of tools without clear guidance on what is approved, appropriate or secure.
At the macro scale, organizations need to treat responsible adoption as an ongoing process rather than a one-off rollout. As AI tools and capabilities continue to evolve rapidly, the guidance around them needs to keep pace.
The organizations creating lasting value from AI will therefore ask very different questions from those focused on token consumption.
Is work becoming more accurate? Are employees making better decisions? Are the most optimal models being used for the type of task? Are we achieving reliable, quality outcomes and optimizing the AI costs to get there? Are repetitive tasks being automated while human expertise is being applied where it creates the greatest value?
When leaders provide the tools, context and confidence employees need, productivity and ROI will naturally follow.
We’ve featured the best AI website builder.
This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.
The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit