40% of AI productivity gains lost to rework for errors

While the promise of enterprise AI use hinges on productivity gains, hidden drains on productivity have emerged in practice, with your organization’s most engaged employees likely suffering the brunt of “AI workslop.”

According to a recent survey from Workday, around 40% of time saved through use of AI is offset by the extra work created fixing AI-generated content. Workday estimates that for every 10 hours of efficiency that companies gain through AI tools, approximately 4 hours are lost fixing AI outputs.

“We’re seeing teams use AI to summarize everything, from simple meeting notes to complex policy or analyst reports. It works for the first, but in the second case, experts often spend more time fixing the output than if they had written it themselves. The real challenge is getting much more granular about where AI adds value — and where it creates rework,” says Laura Stash, executive vice president at iTech AG.

Identifying lost productivity

Leaders need to step back and evaluate where AI adds value, a process that can be as simple as talking to the teams in your organization that are handling the bulk of the company’s AI-generated work, says Paul Farnsworth, president of Dice.

“Look for patterns like certain workflows consistently requiring rework or high performers spending more time editing than creating. AI shouldn’t just accelerate output; it should ultimately reduce friction. If it’s doing the opposite, that’s where leaders need to step in and adjust how this tool is being utilized within their organization,” says Farnsworth.

Lost productivity from AI can quickly become a blind spot for leaders focused on “gross efficiency,” according to Workday. Metrics aimed at the amount of time AI saves can lose sight of the “net value” of AI tools. While there might be early gains in efficiency and speed, it might simply be generating faster outputs — not improving quality or results.

“I’d recommend leaders watch for spots where work starts bouncing back and forth for edits, multiple reviews, or manual fixes, especially from their strongest employees. If speed is up but AI-generated mistakes, revisions, or frustration are also climbing, that may be a sign AI is adding friction instead of value,” says Kareem Osman, VP and market director of technology talent solutions at Robert Half.

Engaged employees

Those most eager to adopt AI are the employees suffering the burden of AI-related rework, Workday found, with 77% of daily users saying they “audit AI work with the same or more rigor than human work.” The bulk of this added labor creates an additional 1.5 weeks of time “lost to fixing AI outputs per highly engaged employee, per year.”

“A company’s strongest employees often become the safety net — they’re the ones catching mistakes, fixing issues, and making sure things don’t slip through the cracks. Over time, that can feel less like high-impact work and more like constant cleanup, which is unsustainable long term,” Farnsworth says.

Stash sees the biggest issues around quality and “cleanup work” with AI arising when employees use it to generate more complex work, especially when they aren’t fully trained on AI tools.

“There’s a place for AI in low-value, repeatable work, but applying that same approach to high-expertise tasks without proper training or validation creates more problems than it solves. If people can’t refine or trust the output, productivity gains disappear — and over time, you risk eroding expertise altogether,” says Stash.

Training

According to Workday’s report, 66% of leaders cite AI skills training as a top investment priority, but only 37% of employees who use AI daily reported increased access to training. This has created an imbalance at many organizations where there’s an expectation to generate high-quality work with AI, without the proper training or skillsets to do so effectively.

Leaders need to align their AI expectations to training and upskilling efforts, Dice’s Farnsworth says.

“That means investing in enablement by training people not just on how to use AI, but how to use it well. It means putting guardrails in place so outputs are reliable and aligned with business goals, and it also means continuously reassessing impact, not assuming that faster automatically means better,” he adds.

It might require redefining job roles, updating employees on new skills they’ll need for their positions, and offer clear guidance to employees on how and when to use AI. The workday report found that 54% of AI users who report struggling with the technology say their required skills haven’t been updated, leaving them unsure of exactly where to start with learning AI skills.

“Employees need clear guidelines on when to use AI, how to validate output, and what success actually looks like. The companies doing this well pair AI with training, quality standards, and accountability, so it helps people do better work, and not just more work faster,” says Robert Half’s Osman.