AI Productivity Gains Remain Elusive for Most Companies

The productivity paradox

Companies are pouring billions into artificial intelligence with the promise of dramatic productivity gains, yet the expected returns remain stubbornly out of reach for most organizations. While individual workers report completing tasks faster with AI assistance, new research reveals a stark disconnect: roughly 90% of firms actively using AI reported the technology had no impact on productivity over the prior three years, according to a February National Bureau of Economic Research working paper surveying nearly 6,000 executives.

The gap between individual efficiency and organizational productivity has created what McKinsey senior partner Alexander Sukharevsky calls a "gen AI paradox." Software engineer Iren Azra Zou at trucking logistics startup Double Nickel says Anthropic's Claude Code helps her complete tasks in a day that previously took a week. Yet these individual wins haven't translated into the company-wide transformation executives anticipated.

Why it matters

The stakes extend beyond disappointed quarterly results. A new Wharton paper warns that if the anticipated productivity boom doesn't materialize, "the current buildout will be the largest misallocation of capital in history," with some major tech companies potentially risking bankruptcy. This assessment comes as companies increasingly cite AI when announcing layoffs or hiring slowdowns, making decisions based on productivity gains that haven't fully emerged.

The automation investment phase

For many workers, AI is actually creating more work in the near term. Amazon data scientist Sarthak Gupta describes working longer hours during what he calls an "automation phase," building pipelines, integrating AI tools, and onboarding existing workflows into new systems. The payoff, he notes, comes from systems that deliver value repeatedly over time rather than one-time speed improvements.

This implementation challenge helps explain why AI gains haven't scaled. At Uber, COO Andrew Macdonald said last month there wasn't a direct correlation between increased AI use and "useful consumer features." The issue has sparked debate over "tokenmaxxing"—workers burning through tokens (the units of text data processed by AI models) without necessarily improving company productivity while racking up substantial bills.

The timeline question

Mark Zandi, chief economist at Moody's, told Business Insider he doesn't expect to see significant AI productivity boosts in economic data until at least the late 2020s or early 2030s. Michael Feroli, chief US economist at JPMorgan, offers a slightly more optimistic view, suggesting that because large language models may require less training than previous technologies, AI-driven productivity gains could materialize in "years, not decades."

The current moment resembles the early days of spreadsheet software. When Lotus 1-2-3 launched in 1983, it radically changed how quickly accountants could work, but spreadsheets didn't become the backbone of the global financial system overnight. AI may follow a similar trajectory—transforming from novel software to procedural backbone over time.

Enrique Dans, a professor of technology and innovation at IE University in Spain, emphasizes that AI "is not a mature tool that you can unpack, plug in, and start redefining your processes." Companies are building new infrastructure on the fly while the fundamental rules of business remain unchanged.

These details were first reported by Business Insider in their analysis of the AI productivity disconnect.