Enterprise AI adoption hits 88% as ROI questions persist

AI has penetrated 88% of organizations worldwide, yet the technology's impact on productivity and profit remains stubbornly difficult to measure — a gap that threatens to widen as companies pour more money into deployment.

It has been roughly 1,200 days since OpenAI released ChatGPT, and the technology has spread faster than almost any enterprise tool in history. McKinsey research shows 88% of organizations now report regular AI use in at least one business function, up from 78% a year earlier. Lenovo's CIO Playbook 2026, based on a survey of 920 executives across Asia Pacific, found 95% of enterprises in Australia and New Zealand plan to increase AI investment this year, with an average expected return of $2.85 for every dollar spent.

Yet the gap between adoption and measurable value is wide. A Wharton study of 801 executives found 75% reported positive returns on AI investments, but the EY 2025 C-suite GenAI Survey showed only 8% of organizations can fully measure and allocate AI-related costs. About half of all AI proofs-of-concept ever reach production, according to Lenovo's data, meaning billions in experimentation spending yields no operational result.

"Saying we're stuck in pilot mode is this outdated idea that's wrong," Ethan Mollick, a professor of management at the Wharton School who studies enterprise AI adoption, said. "I'm talking to companies all the time getting real value out of AI."

The 'jagged frontier' problem

Researchers have coined the term "jagged frontier" to describe AI's uneven capabilities. The models excel at structured tasks such as coding, legal-document review and financial analysis, but struggle with contextual work that requires judgment calls, unwritten rules and institutional knowledge that never makes it into training data.

That ceiling limits what current AI can do across the economy. Daron Acemoglu, an MIT economist and Nobel laureate, said he believes present-day AI tools will have an impact on only a fraction of jobs. "Whether you're a CEO, a manager, a journalist, a professor or a construction worker, I see your skills as beyond what AI can perform," he said.

The structural obstacles extend beyond model limitations. Every company's systems and workflows are different, meaning the data architecture, permissions, guardrails and human oversight required to deploy AI usefully must be built from scratch. Benedict Evans, an independent analyst tracking enterprise AI adoption, said the jagged frontier makes it nearly impossible to predict which use cases will work until after a company has already committed resources.

The human factor slows the curve

Technological hurdles may prove easier to overcome than organizational resistance. Executives face five-year planning cycles, depreciation schedules on recently purchased systems and boards demanding returns. Workers who believe they are training their own replacements have little incentive to cooperate.

"What is being sold is this idea of productivity and efficiency," Kate Brennan, associate director of the AI Now Institute, an AI-policy research center, said. "What that means for the people doing the actual work is rarely part of the conversation."

The instinct at most companies is to use AI to automate parts of existing processes rather than redesign the processes themselves. An insurer handling a fender-bender claim might use AI to speed up paperwork while keeping the same layers of review, rather than letting AI assess damage from customer photos and trigger payment in seconds. That kind of reimagining threatens established hierarchies and routines.

Historical precedent suggests deep transformation takes time. Electricity rewired civilization but took four decades to show up meaningfully in productivity data. The internet needed 10 to 15 years to reshape the economy's foundations. James Landay, co-director of the Stanford Institute for Human-Centered Artificial Intelligence, said AI will likely follow a similar arc. "My sense is more like five to 10 years — not the next two or three," he said.

For investors, the timeline matters. Companies selling AI infrastructure — Nvidia, Microsoft, Amazon — have seen their valuations reflect expectations of rapid enterprise deployment. Nvidia trades at roughly 35x forward earnings, pricing in years of sustained data center growth. If enterprise adoption follows the five-to-10-year path Landay describes, the gap between current valuations and actual revenue realization could widen before it narrows. The boosters are directionally right about where AI is heading. The skeptics are probably right about how long it will take.

This article is for informational purposes only and does not constitute investment advice.