Does AI save time? ActivTrak's productivity data is contradictory

A CIO article by Taryn Plumb examines whether AI saves employees time or adds to their workload. Data from ActivTrak, a productivity-monitoring software company, appears to point in opposite directions, according to the report.

ActivTrak's latest research is based on behavioural data from 9,140 knowledge workers at 467 companies, collected over five months (from 1 April to 25 September 2026). According to its findings, heavy AI users work between 8.42 and 9.08 hours a day, compared with 7.58 hours for non-users. That is up to 90 minutes more per day.

This finding contrasts with the company's previous report, published earlier this year. That study was based on 443 million work hours from 163,638 employees at 1,111 organisations over three years (January 2023 to December 2025). It concluded that the average workday had shortened by around 9 minutes (2%). The article notes that the discrepancy may be partly due to the time period and the sample sizes, which are much smaller in the new study.

On concentration, heavy AI users have focus sessions 22% shorter than light users (37.8 minutes versus 48.2), suggesting more fragmented work. They also switch applications 15% more often than non-users. Sarah Altemus, head of ActivTrak's Productivity Lab, explained that these interruptions add friction to high-value work that requires concentration, critical thinking and sustained attention.

The earlier report had already shown mixed signals. Focus efficiency, meaning the percentage of work time spent on concentrated, uninterrupted activity, fell to a three-year low of 60%, and the average length of focus sessions in 2025 dropped 9% compared with 2023. However, daily productive time rose by 19 minutes (5%) to 6 hours and 36 minutes, and productive sessions grew by 13%. Overall task switching rose by 12%, more than 10 minutes. A note in the publication clarifies that some of these figures are unrelated to AI.

The recent study also compared tools with embedded AI and web-based AI tools. Those using embedded AI switch applications 0.6 times per hour. Those using web-based AI do so between 1.7 and 4.2 times per hour, up to four times more than non-users. ActivTrak calls this additional cost the "swivel-chair tax", the extra effort of jumping from one application to another. In addition, light users spend more active time in the applications specific to their role (81.4% versus 78%). Altemus concludes that using AI more does not necessarily mean working with greater focus.

For Altemus, the underlying issue is how organisations manage the extra capacity that AI may generate. If employees finish tasks faster, managers may assume they have room for more responsibilities. Without clear workload limits, the time saved may end up being redirected into more work, with the burnout that entails.

The article also includes a methodological caveat. Sanchit Vir Gogia, chief analyst at Greyhound Research, acknowledges that ActivTrak offers a useful view of how work moves between applications, but notes that the company sells the capability that produces those measurements. What is recorded depends on what the instrumentation detects. An employee may use AI and still appear in the non-user group. Work on an unmanaged personal device falls outside what is measured. An API or a background agent can complete a task without generating a desktop session. Moreover, passive screen time, in-person meetings logged in calendars and paid hours are different measures. Switching applications is also easier to observe than shifting attention within the same application.

Gogia argues that anyone judging the value of AI should establish what happened after the first task got faster. He warns that an improvement per paid hour can hide effort that does not appear on the timesheet, and that the board should be able to say whose workday improved and whose work made that improvement possible.

The article concludes that companies should remain cautious about vendors' promises of productivity gains, because measuring AI's real impact on work remains very difficult.