6 minBusiness
AI Boosts Individual Productivity but Companies Struggle to See Gains
Workers report significant AI-driven productivity gains, but most companies cannot link the technology to higher earnings or lower costs, according to new research. Business leaders face pressure to redesign workflows and systems to capture value at the organizational level.
Artificial intelligence is making individual workers more productive, but those gains are not translating into measurable improvements for the companies they work for, according to a growing body of research that is forcing business leaders to rethink how they deploy the technology.
Researchers have found that AI typically raises productivity on specific tasks by 15% to 30% in real-world settings, and by 20% to 60% in controlled studies. In a McKinsey survey of more than 1,700 professionals across 97 countries, eight in ten respondents said AI had made them more productive at work.
At the company level, however, the picture is far less encouraging. Only 37% of McKinsey respondents could link AI to any impact on their organization’s earnings before interest and taxes, a figure essentially unchanged from a year earlier. The National Bureau of Economic Research reports that while 69% of firms were using AI, 89% of executives saw no impact on productivity over the previous three years. And in PwC’s 2026 Global CEO Survey of more than 4,400 chief executives in 95 countries, 56% said AI had delivered neither higher revenue nor lower costs in the past year.
The disconnect echoes a famous observation by Nobel laureate economist Robert Solow in 1987, when he noted that the computer age was visible everywhere except in productivity statistics. Nearly four decades later, AI has produced its own version of that paradox.
Closing the gap between individual and organizational gains is now one of the most important tasks facing business leaders. The solution, according to analysts and executives who have studied the problem, lies not in the tools themselves but in the systems companies build around them.
History offers a precedent. Economic historian Paul David argued that electricity went through a similar phase. When factories first electrified, owners typically attached electric motors to existing shaft-and-belt systems, often keeping old steam engines in place. The layout of the factory stayed the same, and productivity growth remained modest. The breakthrough came when manufacturers redesigned the factory itself, giving each machine its own motor and building lighter, single-story plants organized around the flow of work. By David’s estimate, the spread of factory electric motors was associated with roughly half of the sharp acceleration in manufacturing productivity growth in the 1920s.
Most companies adopting AI today are still at the bolt-on stage, applying the technology to existing processes without rethinking how work gets done. To capture real value, companies need what some analysts call an abundance system, built on three principles.
The first is to focus on what AI makes possible, not just what it can save. In Deloitte’s 2026 State of AI in the Enterprise survey of more than 3,200 leaders in 24 countries, 66% reported efficiency and productivity gains, but only 20% said AI was helping them improve products and services or innovate. Savings have a ceiling, since you can only cut what you already spend. The bigger prize is work that was never viable before. One team recently built a suite of apps over six months that would have cost roughly five times as much and taken about two years longer without AI, economics that would have made the project nonviable from the start.
The second principle is to build reusable components. Good software engineering means never writing the same code twice, but AI pulls organizations in the opposite direction. Because generating something new is cheap, people start from scratch every time, producing enormous volume but nothing that compounds. Unilever offers an alternative: the company has created digital twins of its products, AI-driven 3D replicas that hold every variant, label, packaging format, and language version in a single file. Each twin is built once and then used to generate imagery for every sales and marketing channel.
The third principle is to orchestrate resources for the AI workflow. When production multiplies, pressure shifts downstream to reviewing, testing, approving, and fitting pieces together. Organizations that fail to redesign for this shift end up with more unfinished work. Software development shows the pattern clearly. After GitHub Copilot’s launch, a study of open-source projects found that output rose, but so did rework, with the burden falling on the most experienced developers, who reviewed 6.5% more code while their own coding output fell.
For business leaders, the message is that AI abundance will not arrive automatically with the tools. It requires deliberate redesign of processes, roles, and organizational structures to ensure that individual productivity gains add up to something larger.
7
