Looking for a productivity breakthrough

2025 was a good year for productivity growth in Britain and America. Both countries combined decent growth with virtually no change in the size of their workforce. The resulting rates of productivity growth – a measure of the efficiency of production – were roughly twice normal levels in the US and UK last year.

This may sound a bit obscure, but for economists it is thrilling. Getting more output from the same number of people is the Holy Grail of economic policy. Faster productivity drives material progress, raising living standards, creating better jobs and improving public services. Every government wants to raise productivity growth. Few achieve it.

Last year’s productivity data have ignited a debate among economists. Could AI explain the rise in productivity growth?

We are not yet convinced. There’s a question of causation. Technology isn’t the only factor that affects productivity. Shifts in the composition of the workforce matter too. The Trump administration’s clampdown on immigration has significantly reduced the supply of less-skilled workers. Arithmetically, that boosts measured productivity, even as it depresses output and employment. More generally, productivity data are notoriously erratic. One good year is not a trend.

This is not to dismiss the potential of AI. Previous technologies have taken years, sometimes decades, to show up in the whole-economy productivity data. Those data are too broad-based and lagging to pick up early signs of structural change driven by new technologies. Change proceeds at different rates depending on the scale of the organisation and sector. To map this sort of granular change we need industry studies. Here, things get more interesting.

A Bank for International Settlements study published in January matched survey data with financial statements from 12,000 non-financial firms across the EU and found that AI adoption raises labour productivity by 4%. It may not sound much, but for labour productivity a 4% gain is a lot. The gains were due to better use of capital, with, perhaps surprisingly, no negative effects on jobs.

There have also been a number of other studies and reports showing positive effects from AI-adoption, albeit in specific or specialist applications.

A Swedish trial used AI-assisted mammography screening to achieve a 29% improvement in detection rates. A Fortune 500 software company raised output in customer support by 14% using generative AI. Bank of America’s digital assistant ‘Erica’ has reduced call centre volumes by 40%.

AI may not be showing up in the high-level productivity data, but we can see some evidence of more rapid change in some parts of the economy. To translate this into a step change in whole-economy productivity would, at the very least, require higher levels of AI use and the reorganisation of processes to realise the potential of AI.

History shows that the hardest part of innovation is often the prosaic part – not invention, but mass adoption, the reorganisation of the workplace and reskilling.

The West desperately needs a new productivity-boosting technology. Whether AI is that technology is one of the great economic issues of our age.