AI Skills: The Budget’s Missing £54 Billion

The Budget came out on Wednesday. Growth is supposedly the government’s number one priority.

So I spent Thursday morning going through the documents, looking for the stuff that would actually move the needle on productivity. The infrastructure spending, the planning reforms, the skills investment.

And I found almost nothing.

Which is frustrating, because there’s one opportunity sitting right in front of them that could add £54 billion a year to UK GDP. It’s not complicated. It’s not politically contentious. And it could start tomorrow.

AI training for the existing workforce.

Here’s what we know

We train people to use AI. Thousands of people, across hundreds of organisations, from private equity deal teams to healthcare communications managers. And we track what happens afterwards.

The pattern is remarkably consistent. After training, 80% of participants go on to use AI daily in their work and they save an average of 47 minutes a day.

I got curious about what those numbers would mean at a national scale.

The national calculation

The UK has about 34 million people in employment. ONS occupation data shows roughly 46% work in managerial, professional, associate professional, and administrative roles (the jobs where AI tools are most useful). That’s about 15 million workers.

If you trained all 15 million you’d spend £5.2 billion. If 80% of them save 47 minutes a day over 230 working days, that’s 180 hours per person per year. Value that at £25 per hour (median wage plus employer costs), and each daily user generates about £4,500 in annual productivity.

Multiply it out and you get £54 billion. Every year. From a one-off training investment.

That’s an increase of 1.9% of GDP. The return in year one alone is roughly 10:1.

Checking my working

Now, you might reasonably think I’m being optimistic. I run a training company. Of course I think training works.

So I checked our numbers against research from people who have no stake in selling training.

Anthropic (the company behind Claude) published their Economic Index earlier this year. They analysed 100,000 real AI conversations to estimate productivity impacts. Their finding: AI could add 1.8 percentage points to annual US productivity growth with broad adoption. That’s almost exactly what our data suggests.

The academic studies show the same direction at the individual level. Harvard and BCG tested 758 consultants on real tasks and found those using GPT-4 completed 12% more work, 25% faster, at 40% higher quality. Stanford and MIT studied 5,179 customer service agents and found 14% average productivity gains (35% for novices). Peng et al. found software developers using GitHub Copilot increased output by 26%.

And there's direct evidence that training specifically makes the difference. The London School of Economics and Protiviti surveyed 3,250 workers globally and found that employees with AI training were twice as productive with AI tools as those without.

Different researchers, different methods, different contexts. All showing that AI makes trained workers meaningfully more productive.

Why this matters more than usual

The UK has a productivity problem. It’s been going on for years and nothing seems to fix it.

ONS figures show the US produces about 28% more output per hour than we do. France and Germany are 13-14% ahead. The London School of Economics have shown that about half of this gap comes down to underinvestment in capital, with slower skills growth also contributing. Not regulation. Not culture. Investment.

And we keep falling further behind. CIPD and Eurostat data show UK employers spend roughly half as much on training per employee as Germany, France, or the Netherlands. The gap has widened since 2005.

Training works. This isn’t controversial. An academic study tracked British industries for 13 years and found that raising the proportion of trained workers by one percentage point increased productivity by 0.6%. Studies from Portugal and Belgium found gains of 2-10%.

The evidence has been there for decades. The UK just doesn’t act on it.

What AI changes

General training takes years to show results. AI training is different. The returns come in weeks.

A UCL study from August found 7.4 million UK workers were already using AI tools by mid-2024. But most had no training. Microsoft’s Work Trend Index found 75% of knowledge workers use AI, but only 39% have had any employer training.

That gap between using AI and using it well is where productivity is being lost.

The same UCL study also found some uncomfortable things. Job postings in AI-exposed roles fell 5.5% compared to pre-ChatGPT trends. The wage premium for AI-exposed tasks dropped 12% between 2017 and 2024.

What’s actually in the Budget

The Budget has £1.5 billion for the Youth Guarantee and Growth and Skills Levy, which funds apprenticeships and helps young people into work. That’s fine. But it does nothing for people already in jobs who need new skills.

The Growth and Skills Levy launching in April 2026 broadens what apprenticeship funds can cover. But it still focuses on qualifications. A marketing manager who wants to learn AI for their current job doesn’t need a certificate. They need practice.

The Budget had money for AI research. Money for AI infrastructure. Money for AI safety.

But nothing for helping workers actually use AI.

What would actually help

The government doesn’t even need to spend the £5.2 billion itself. By doing a couple of simple rule changes they could get firms to deliver this training as part of their existing plans.

Expand the Growth and Skills Levy to cover practical skills training, not just apprenticeships. Apprenticeships are designed to help people move into new roles, not to help them do their existing job better. A marketing manager who wants to learn AI doesn’t need a 12-month programme with a qualification at the end. They need a day of hands-on practice and some follow-up support.

Make it mandatory for firms to provide all workers with a minimum of four days training a year. This costs the government nothing and eliminates the free-rider problem (where companies underinvest in training because they’re worried staff will leave and take their skills elsewhere).

The bottom line

Training 15 million workers in how to use AI costs £5.2 billion. Based on our data, that generates £54 billion a year in productivity. Even if you halve the estimates for caution, the return is compelling.

The government says growth is the priority. Here’s an easy win with strong evidence and fast returns.

I’m genuinely not sure why they’re not taking it.