AI projects not paying off? Keep at it, says Uni Productivity Institute
A University of Manchester research institute says businesses should give AI investments time to produce returns, as early costs and faster individual tasks have not yet led to broad productivity gains. Its report examines why AI adoption has fallen short of expectations and sets out steps for businesses and policymakers to help close what it calls the “AI ambition gap”.
The report says spending on staff retraining and work changes is immediate, while benefits may take longer to emerge. It cites studies in which developers completed 26 per cent more tasks with GitHub Copilot and professionals using ChatGPT cut routine task time by 40 per cent while improving output quality by 18 per cent; however, AI coding tools raised coding activity by about 240 per cent but software releases by only 30 per cent. The UK has the highest share of workers in highly AI-exposed occupations with strong human complementarity among six countries assessed, and could see annual labour productivity gains of 0.4 to 1.3 percentage points, according to OECD estimates cited in the report.
In short
- AI investment costs arrive before productivity gains.
- Faster tasks do not always mean more output.
- The report urges stronger management, skills and AI governance.
Steelman: both sides, in good faith
The strongest fair case each way — we don't pick a winner.
The case for
Businesses should give AI adoption time because retraining, redesigning workflows and integrating tools impose costs before benefits can spread across an organisation. Studies cited in the report suggest AI can help people complete routine tasks faster or produce more, and the UK’s high share of workers in AI-exposed roles could support substantial productivity gains if workers’ skills complement the technology.
The case against
Businesses should be cautious about sustaining AI investment when faster individual tasks have not yet translated into comparable gains in organisational output. The reported gap between a sharp rise in coding activity and a much smaller increase in software releases suggests that efficiency claims may overstate practical returns, while training and workflow changes carry real costs and future gains remain uncertain.