AI’s productivity dividend is finally showing up on the books: here’s where, and how to claim yours
For two years the AI conversation has run on promise. This month it ran on numbers. A cluster of UK reports and rollouts has put hard figures against a question every leadership team has been asking: is artificial intelligence actually making work better, or just busier? The answer, increasingly, is the former, but only for organisations that treat AI as a change programme rather than a software purchase.
We’ve pulled together the most credible evidence published recently, and the pattern is striking. The gains are real, they’re measurable, and they’re landing first in exactly the kind of regulated, knowledge-intensive settings our clients operate in.
The proof points are getting hard to ignore
Start with the largest example in the country. NHS England is rolling out Microsoft 365 Copilot to more than 505,000 staff after a trial spanning 30,000 workers across 90 organisations, the biggest healthcare AI trial of its kind globally. The measured result: an average of 43 minutes saved per person per day, equivalent to roughly two working days a month, or five weeks a year. Separate AI scribe trials in London delivered 23.5% more direct patient contact time and 13.4% higher A&E throughput per shift. These are not vendor projections. They’re outcomes from a live trial at national scale.
The professional services world is producing equally concrete figures. In the UK legal sector, more than half of firms now use AI-assisted document review; 43% of solicitors report improved productivity and work quality, and tellingly, 54% now measure AI success by time saved, a sign the technology has moved from curiosity to managed capability. Small firms deploying automated document tools report 30% higher client retention. In accountancy, professionals are recovering an average of 5.4 hours a week, with advanced users saving 71% more time than beginners and reinvesting it into advisory work rather than data entry.
Zoom out and the cross-sector picture holds. Lloyds’ latest business barometer found 82% of AI-using firms reported higher productivity and 76% higher profitability, with retailers seeing the strongest gains. The London School of Economics put a memorable figure on it: AI is worth around one working day per week to the people using it well.
But the headline number isn’t the whole story
Here’s where we’d urge a degree of discipline. The same week’s evidence carries a warning that’s easy to skim past. UK workers may save around 12 hours a week with AI, but research suggests 6.3 of those hours go straight back into “botsitting”: checking outputs, correcting errors, re-prompting. Only 18% of organisations say AI has significantly improved their overall performance. And in the legal sector, AI-generated citation errors have already produced referrals to the Solicitors Regulation Authority.
What this tells us is that the dividend is not automatic. The Skills England “What Works for AI Upskilling” report, drawing on more than 150 employers, found that the standout results came where training and workflow redesign accompanied the technology: Roche saving up to four hours per user per week, KPMG saving time across 17,000 staff, and a marketing agency adding an estimated 20% in new AI-enabled revenue.
Our take
We think 2026 is the year the question changes. It’s no longer “does AI work?” The sector evidence has answered that. The question is “why isn’t it working here?” And for most organisations, the honest answer is that they’ve bought licences without building the capability, the process changes, and the verification habits that convert saved minutes into genuine value.
That gap, between time saved and value captured, is precisely where we focus. The firms pulling ahead are doing three unglamorous things well: they measure a baseline before they deploy, so they can prove the gain; they redesign the workflow rather than bolting AI onto an existing one; and they build verification in by design, so quality and compliance keep pace with speed. None of that is exotic. All of it is the difference between a press release and a P&L impact.
Where to start
If you want this quarter’s AI spend to show up on next quarter’s numbers, three practical steps will take you most of the way:
- Measure the baseline. Pick one high-volume, knowledge-heavy process (document review, reporting, client correspondence) and measure how long it takes today. If you can’t measure it, you can’t prove the gain.
- Invest in capability, not just access. The data is unambiguous: trained users dramatically outperform untrained ones on the same tools.
- Design the checking step before you scale. Especially in regulated work, where an unverified output is a liability, not a saving. To put indicative numbers against your own operation, try the AI ROI Calculator.
At Boxtree Consulting we help professional services firms and growing businesses do exactly this, turning AI from a line of cost into a measurable operating advantage, with the governance to keep it safe. If you’d like to understand where the dividend is hiding in your own operations, we’d welcome a conversation.
The productivity is there. The proof is now public. The only question left is whether your organisation is set up to claim it.
See where the dividend is hiding in your operations
Boxtree finds where AI can release capacity and cut cost across your operations, then builds and embeds it, with the governance to keep it safe.
