The Productivity Paradox: What AI Adoption Data Reveals

In 1987, Robert Solow observed that although businesses seemed to be entering the computer era, productivity figures did not seem to support this. He was discussing a period when businesses were investing heavily in computers, but it was still difficult to establish quantifiable productivity benefits. The productivity paradox is the name given to this discrepancy between investment and outcomes. Nearly 40 years later, UK businesses are seeing what appears to be a repetition as AI adoption is increasing.

Adoption Is Rising, but Not Deepening

The percentage of companies in the UK that employ at least 10 employees and currently use some kind of AI tool grew from 12 per cent to 35 per cent between late 2023 and June 2026 according to the Office for National Statistics. At face value this is quite a quick transformation. However, they did not seem to be introducing a wider range of AI tools. The average number of AI tools used by adopters changed from 1.4 in 2023 to 1.6 in 2026, and while companies were using AI, they were not using a wide range of them.

That growth was not evenly distributed either. AI is used by 58% in the information and communication businesses, compared to 13% for construction. Firm size is also a factor, with separate DSIT research finding 36% of large firms had adopted AI compared with just 14% of micro-businesses. Consequently, the adoption of AI feels less like a wave washing over the economy and more like pockets of AI use in specific industries and larger companies. 

Why the Gap Persists

The same DSIT research was conducted in 2025 and found that 75 per cent of businesses using AI had experienced productivity gains, but only 12 per cent said they had seen an increase in revenue. This means AI in its early phases can improve efficiency in internal processes, without necessarily altering what a business can charge or sell. The impact of AI adoption on the labor market is also similarly patchy, with effects on employment and wages proving elusive early in the cycle. Skills gaps were identified as the most common barrier (60% of businesses surveyed), followed by cost or trust. For firms weighing whether to invest further, the opportunity cost of retraining staff or restructuring workflows around AI tools competes directly with other uses of the same time and capital. That comparison does not always favour further AI investment.

Why the Easy Wins Come First 

Where adoption has taken hold, it has concentrated in a narrow set of applications. Large language models are the most widely used category among UK businesses, followed by visual content creation tools, including platforms like Seedance 2.5 on Pollo AI, which allows businesses to generate and adapt short marketing or product videos without commissioning a new shoot for each variant. These are low-cost, quick-turnaround applications that a firm can adopt without restructuring how it operates. This is consistent with the law of diminishing returns.

Has the Paradox Happened Before? 

Eventually, in the 1990s, Solow's paradox was solved, as businesses spent a decade re-engineering their processes around the technology, instead of just layering it on top of existing ones. This is a reflection of the bigger question of what causes productivity growth at the national level, where investment alone has seldom been enough to move the numbers without changes in how that investment is used. The AI data today suggest a similar lag: narrowly measured, adoption looks transformative, but the productivity and revenue effects that would confirm a true shift have not yet appeared at the same scale.