In 1975, a Bank of England economist named Charles Goodhart noticed something odd. The U.K. government had started targeting specific money supply indicators as proxies to control inflation, a notoriously complex thing to measure directly. But like a performance review that incentivizes quantity over quality, banks changed their behavior to game those exact metrics. As a result, the money supply looked like it was under control, and the government kept tracking the numbers. And yet inflation persisted. Economists, with their usual flair for naming things, turned this into Goodhart’s Law: When a measure becomes a target, it ceases to be a good measure.
Today, something similar is happening with AI. Organizations are tracking adoption: who’s prompting, how often and how much employees are using AI. Some have gone further, tying performance reviews and even bonuses to how much AI their people use, a.k.a “tokenmaxxing.” And just like the banks in 1975, people are responding rationally to the incentive. Usage is up. But whether that usage is producing anything worth having is a question most dashboards can’t answer.
What Metrics Really Matter for AI Adoption?
- Where saved time goes: Redirecting freed hours into strategic planning and employee development increases team AI performance by 65 percent, yet only 7 percent of organizations guide employees on reinvesting saved time.
- Impact on human relationships: Using AI to avoid direct check-ins increases team burnout by 26 percent and intent to leave by 29 percent. Using AI to prepare for check-ins improves team connection.
- Employee trust: Employees who trust leadership are 46 percent more likely to view AI as augmentative (enhancing their value) rather than automative (replacing them).
The Enablement Illusion in Action
Gartner surveyed 12,000 employees and managers across 40 countries earlier this year and concluded that most leaders are confusing adoption metrics with actual progress. They called it the “enablement illusion.” In our own research on AI readiness, we found a similar pattern. Those with equally high adoption scores can vastly differ in outcomes ranging from well-being to productivity, depending on if you take into account a variety of human behaviors and skills, traditionally characteristic of good management. Put another way, adoption scores mask a lot of heterogeneity.
The difference comes down to two main things: the ways people are using AI and the organizational conditions that shape those behaviors. For individuals, what matters is how managers spend the time AI saves them, whether they maintain or reduce human relationships and if they approach AI with a sense of agency as opposed to compliance or threat.
Organizationally, those individual behaviors are afforded by the conditions leadership builds. For example, does management culture support trust, feedback and development? Is leadership communication about AI clear, and does it carry a credible commitment to augmentation over automation? Has the organization invested in the agency of the talent infrastructure through training, policies and AI-related KPIs?
1. What Are Employees Doing With Saved Time?
Similar to other global industry reports, our study found that AI saves managers about six hours a week. How those saved hours were reallocated is a strong predictor of team AI performance. Most managers put the time right back into more of the same: improving the quality of existing work or doing more admin. Managers who redirected time into developing their people, their own professional growth, new projects and strategic planning saw a 65 percent increase in team AI performance, however.
According to Gartner, only 7 percent of organizations provide any guidance to employees on what to do with AI-saved time. That gap between freed capacity and intentional reinvestment is where a lot of AI value is leaking. If managers don’t have guidance on how to use the hours, they default to doing more of the same.
2. Is AI Building Relationships or Replacing Them?
Some managers use AI to avoid important conversations like developmental check-ins, career guidance and direct feedback. People can easily slip into that mode when squeezed with pressure to move fast, but the consequences can’t be ignored. Our research shows team coordination drops 12 percent, burnout rises 26 percent and intent to leave jumps 29 percent.
When managers use AI to prepare for those conversations, they report being more prepared, confident and less anxious. Further, when in an environment of high psychological safety, if managers use AI in that preparatory way, cross-functional relationships increase over the following year and people report stronger belonging and connection to their teams.
Track whether rising AI usage correlates with more human interaction or less. If your managers are having fewer direct conversations as their adoption scores climb, the coordination and retention numbers will follow, even if they don’t show up for another quarter.
3. Do Your People Trust the Reason AI Is There?
Our research shows that employees who trust leadership have 46 percent higher odds of experiencing their organizational motivation with AI as augmentative, where the technology increases their capabilities and value, rather than automative, which replaces them. Our research suggests companies who signal and encourage a perception of augmentation are more likely to be successful in the long run, maintaining higher retention, productivity and a robust leadership pipeline. If you haven’t directly asked your employees whether they believe AI is there to augment them or replace them and whether they trust their leaders, they’ve already answered that question for themselves.
Measure the Right Thing
Goodhart watched the Bank of England learn his own law the hard way: once you target a number, people optimize for that number and it stops meaning anything. Right now, most organizations are targeting AI adoption. The predictable result is a lot of adoption and very little clarity on whether it’s helping. The three things worth measuring — where freed time goes, what’s happening to human relationships and whether your people trust the reason AI is there — won’t appear on your AI dashboard. But they will appear in your long-term results.
