The numbers come first.
ActivTrak's Productivity Lab tracked 120,620 employees across 1,009 organizations for three consecutive quarters — Q4 2025 through Q2 2026 — and the findings challenge one of the most expensive assumptions in corporate America right now: that maximum AI adoption equals maximum productivity.
According to Heidi Farris, CEO of ActivTrak, the data maps workers into three behavioral stages. Stage 1 — Research Assistance — covers employees who use AI like a search engine to answer questions and summarize information. Stage 2 — Task Execution — covers those who draft content, generate ideas and complete routine tasks that they then validate and finalize. Stage 3 — Workflow Integration — is where AI becomes an integral part of day-to-day operations.
The breakdown: 27% of the employees studied fell into Stage 1, 14% into Stage 2, and only 2% reached Stage 3. Overall, AI users remain a minority at 43% of the workforce studied.
Where productivity actually peaks
Healthy utilization — the metric ActivTrak uses to capture both productivity and work-health indicators — rises as employees move from little or no AI use toward regular, task-level adoption, peaking at 75%. But once AI becomes embedded in workflows at Stage 3, healthy utilization drops approximately 5 percentage points, falling to levels statistically indistinguishable from employees who barely use AI at all.
Stage 2 is where the real operational gain lives, Farris writes. A sales representative generating a quote from five systems with a single prompt instead of manually compiling the information is the kind of friction elimination that moves the needle. That is where most organizations should be focused, according to the ActivTrak data.
Two risks of chasing the deepest tier
Farris identifies two specific failure modes when organizations push past the optimal adoption point.
First, runaway costs. More mature usage means more powerful models, more tokens and more infrastructure. If the task or role does not require that level of sophistication, the organization is spending capital it could deploy elsewhere. ActivTrak's own operations team discovered employees routinely using the newest, most powerful Anthropic model to rewrite customer emails — a task that did not require it. The company responded by creating an internal reference to help employees match the right model to the right task.
Second, operational disconnect. Employees who sprint ahead may generate sophisticated AI workflows that optimize individual tasks without improving broader processes. Without workflow redesign tied to business goals, Farris writes, 'all you've done is produce more AI slop, faster.'
Adoption is sticky — which makes the call consequential
The stickiness of AI behavior makes the leadership decision durable. ActivTrak's data shows 82% of employees who adopted AI kept using it. Once workers move past casual use, they continue quarter after quarter, and almost no one who goes deep ever reverses course. As Farris puts it: 'The level of adoption I push my team toward is the level of adoption they'll likely stick with.'
The prescription is straightforward: map workflows before deploying tools, resist department-wide mandates in favor of targeted fixes, and build measurement into the process from the start.
CEO Times take: Free enterprise runs on capital allocation, and the AI arms race is no exception to that discipline. The ActivTrak data is a useful corrective to the vendor-driven narrative that more compute always means more output. Matching the right tool to the right task is not a technology problem — it is a management problem, and the organizations that treat it as one will compound their productivity gains while competitors burn budget chasing a maturity score that does not translate to earnings.
