Study of 120,620 workers shows peak AI gains at moderate use, not maximum

ActivTrak's three-quarter study challenges the prevailing assumption in AI adoption strategy that more AI use always yields better results. The behavioral data from 120,620 employees across 1,009 organizations reveals a counterintuitive pattern: productivity rises as employees progress from no AI use through regular, task-level adoption, but then plateaus and declines once AI becomes embedded in workflows. This suggests that the conventional AI maturity models — which typically measure consumption metrics like licenses or login counts — are measuring the wrong thing and may be steering organizations toward overconsumption.

The research identifies two concrete operational risks of pursuing maximum adoption without intentionality. First, organizations may accumulate runaway costs by deploying powerful models to tasks that do not require them; ActivTrak's own operations team discovered employees routinely using the newest model to rewrite customer emails, a task that needed far less sophistication. Second, employees may optimize individual workflows without regard to broader business processes, generating what the CEO calls "AI slop, faster" — sophisticated outputs that do not improve organizational outcomes. The durability of adoption (82% retention, and almost no reversion from deep usage) means that leaders who push teams toward maximum adoption may lock in behaviors that no longer serve the business.