To measure the ROI of AI, take ‘workslop’ into account

With so many CEOs doubling down on AI initiatives, AI is poised for a consequential year in 2026. So consequential, in fact, that half of CEOs, according to the Boston Consulting Group’s January 2026 AI Radar survey, believe that their jobs are dependent on “getting AI right.”

But studies provide conflicting data about the ROI of AI.

A December 2025 EY pulse survey of 500 business leaders, for instance, found that 96% of organizations have seen productivity gains from AI, with more than half of respondents (57%) describing the gains as “significant.” But more than half of the 4,454 global CEOs in a survey fielded by PwC in fall 2025 said that AI hasn’t either lowered costs or increased revenues.

One method of tracking AI’s impact on productivity is in terms of time savings. A 2025 survey by the London School of Economics found that AI use can save employees an average of 7.5 hours a week, for instance, while a 2025 MIT Sloan and Stanford University study estimated that accountants using AI could shave 7.5 days off the month-end close.

Time saved, however, isn’t the only metric CFOs need to track AI productivity gains. The impact of “workslop,” or low-quality AI-generated output that employees must spend time correcting, should also be considered.

A November 2025 survey by Workday found that more than a third of the time that employees save using AI is negated by rework. Though 85% of employees said that AI saves them one to seven hours per week, they reported spending 37% of this time correcting, rewriting, or contextualizing subpar AI output. That translates to 1.5 weeks a year, the authors of a report on the survey said.

Workday polled 3,200 company leaders and full-time workers at organizations with at least $100 million in annual revenue and 150 or more employees. Respondents were spread among companies in EMEA, APAC, and North America and evenly split between leaders and employees.

Workday’s findings are consistent with those from a survey by Stanford University and BetterUp Labs, which found that 40% of US employees had received workslop from a coworker in the previous month. In addition, survey respondents said about 15% of the work product they review is workslop, and they spend about two hours dealing with each instance.

Solid AI policies and proper tone at the top can help reduce the incidence of workslop, experts told CFO Brew in November. Workday recommends companies measure the “net value” rather than the “gross value” of AI-related productivity by including rework in their calculations.

But that “net value” might be lower than expected: “Only 14% of employees consistently achieve net-positive outcomes from AI use,” the survey’s authors said.