The AI productivity paradox

Two different stories

Ask a C-suite executive how much time AI saves them and you'll get a confident answer. In a survey of 5,000 knowledge workers last fall, AI consulting firm Section found that more than 40% of executives said AI saves them over eight hours per week. Nearly one in five claimed more than twelve. Only 2% said it saves them nothing.

Ask the people who report to those executives and the story inverts. Forty percent of non-management workers said AI saves them no time at all. Two-thirds reported saving less than two hours a week, or nothing. Just 2% said it saves them more than twelve hours.

Same companies. Same tools. Completely different realities.

The financials that should follow

If AI were delivering the productivity gains that executives experience, you'd expect to see it in the financials. You don't.

PwC surveyed 4,454 chief executives across 95 countries for their 29th Annual Global CEO Survey, released today ahead of the World Economic Forum's annual meeting in Davos. Fifty-six percent said AI has failed to either boost revenue or lower costs. Only 10% to 12% reported seeing benefits on either side. PwC's global chairman, Mohamed Kande, attributed the gap not to the technology itself but to a lack of foundational rigor: "Somehow AI moves so fast that people forgot that the adoption of technology, you have to go to the basics." Clean data. Solid business processes. Governance. The companies seeing results are the ones that put the foundations in place before layering AI on top.

A frequently cited MIT study, "The GenAI Divide," found the same pattern from a different angle: 95% of companies attempting to incorporate generative AI saw no meaningful revenue growth.

These are not marginal findings from peripheral researchers. PwC and MIT are about as mainstream as institutional credibility gets. The message is consistent: for most companies, the AI investment isn't translating into financial results.

Why the boardroom experience is real

The executives reporting significant time savings aren't making it up. Their work maps directly onto what AI does well.

Executive days are filled with drafting communications, summarizing long documents, synthesizing information across sources, preparing for meetings. These are high-frequency, language-heavy tasks. They're exactly what large language models were designed for.

Section's data confirms it: 57% of C-suite respondents use AI daily, and 94% say they trust it. At that level, the productivity gains are real because the work being augmented is the work AI is best at augmenting.

The mistake is assuming that experience generalizes across the organization. It doesn't.

Why the floor experience is also real

For the people doing the frontline work, AI introduces a cost that never shows up in the boardroom metrics.

Workday surveyed 3,200 employees at companies with over $100 million in revenue and found that 37% of the time saved through AI is immediately consumed by checking, correcting, and rewriting AI output. For every ten hours of efficiency gained, nearly four are lost to rework.

Think of it as an AI tax. It's real, it's measurable, and it falls almost entirely on the people doing the hands-on work. For frequent AI users, Workday estimates the rework adds up to about a week and a half per year. The researchers calculated an invisible cost of roughly $186 per person per month. Not a technology line item. A human cost, paid in attention and effort by the people who can least afford to lose either.

Seventy-seven percent of daily AI users told Workday they scrutinize AI output as carefully as they would a colleague's work. If you have to verify everything AI produces with the same rigor you'd apply to a human draft, the efficiency story starts looking different from the one being told upstairs.

The proficiency gap that widens the tax

The Section survey reveals another layer. Only 3% of individual contributors qualify as what Section calls "proficient practitioners." Seventy percent are basic experimenters who use AI for simple tasks like rewriting emails and getting quick answers.

The gap in outcomes is enormous. Proficient practitioners are 20 times more likely to save four or more hours per week. But proficiency doesn't come from enthusiasm or mandate. It comes from training and workflows redesigned around what the tools can actually do. Most organizations haven't provided either.

Sixty-six percent of leaders say skills training is a top priority. Only 37% of employees facing the most AI rework say they're actually receiving it. The training gap is widest exactly where it matters most. Workday described the broader structural problem as "forcing employees to use 2025 tools within 2015 job structures." The AI tax isn't just a technology problem. It's a design problem. And most organizations haven't redesigned anything.

The emotional data tells the same story from a different direction. Two-thirds of non-management workers in the Section survey said they feel anxious or overwhelmed about AI. Less than half of managers felt the same. Nearly 75% of executives said they're excited. The people closest to the friction are the most stressed by it. The people furthest from it are the most optimistic.

The Solow echo

The gap between technology adoption and productivity growth has a name. Economist Robert Solow observed in 1987 that "the computer age can be seen everywhere except in the productivity statistics." Analysts call it the Solow Paradox, and the historical data supports the pattern.

Before personal computers, U.S. productivity grew at 2.7% annually. After PCs went mainstream, it dropped to 2.1%. After smartphones, 1.5%. In 2024, with AI investment at all-time highs, U.S. productivity growth showed what Forrester VP and principal analyst JP Gownder calls "no clear AI signal."

New technology does not automatically produce productivity gains. It never has. The gains come later, after organizations restructure work around what the technology makes possible rather than layering it on top of what already exists.

The split that matters

Anthropic's latest Economic Index, published earlier this month, adds a useful complication. Forty-four percent of jobs can now use AI for at least a quarter of their tasks, up from 36% in the prior report. AI adoption is spreading faster across the American workforce than any major technology in the past century.

But the impact splits along a line worth paying attention to. Some roles see real augmentation. Radiologists spend less time on image analysis and more time with patients. Therapists offload documentation and gain clinical hours. In these cases, the freed time has somewhere valuable to go, and the AI tax is low because the outputs serve as inputs to human judgment, not as finished work.

Other roles are simply deskilled. Data entry workers, IT specialists, travel agents. The work gets easier without creating any higher-level work to fill the gap. The AI tax in these roles is high because the outputs require heavy correction, and "easier" isn't the same thing as "more productive" when there's nothing better to do with the difference.

The honest count

The AI productivity paradox is not a story about technology failing. It's a story about measurement and design.

Executives are measuring their own experience. That experience is genuine. AI saves them significant time because their work is the kind of work AI was built to handle. Employees are measuring their own experience too. That experience is also genuine. It includes rework, unreliable outputs, missing training, and job structures built for a different era.

At the company level, 56% of CEOs report no financial benefit. Ninety-five percent of companies see no revenue growth from AI. The executive experience, however real, isn't representative of what's happening across the organization.

The starting point for any honest AI strategy is the distance between these two stories. Not the optimistic version. Not the pessimistic one. The gap itself. Because the companies that take that gap seriously will eventually find real productivity. And the companies that mistake one floor's experience for the whole building's will keep waiting for results that never arrive.