AI training for business: what a company is actually buying

The tools arrived before the skills

The Office for National Statistics publishes the cleanest read on this. Its Business Insights and Conditions Survey ran from 5 to 28 June 2026 with 38,637 responding businesses, and found around 35% of businesses with 10 or more employees using at least one AI technology, up from roughly 12% in late 2023. Among businesses with 250 or more employees it was 49%. The most common single use was text generation with large language models, at 18%.

Then the number that should change how you budget. In the same survey, only 11% of businesses with 10 or more employees reported that more than half their workforce had received AI related training.

So the median UK business has the software and has not trained the people. That is not really a training problem yet. It is a scope problem, because nobody has decided who needs to be good at what.

Why everyone says they train and almost nobody has

In June 2026 the SKAI programme published What Works for AI Upskilling in the UK, written by Dr Nisreen Ameen at Royal Holloway under a British Academy Policy-Led Innovation Fellowship in partnership with Skills England. It draws on 23 workshops, ten case studies and a survey of 536 responses, which makes it the most specific UK evidence base on this question currently available.

Nearly all organisations in that survey, 97%, reported providing AI training. The same respondents then named the gaps. 51% flagged flexibility, 35% the absence of a clear AI skills framework, 34% practical and contextualised learning, 29% ethics and governance, and 22% weak leadership buy in.

Read the two surveys together and the picture resolves. Most businesses have bought something they can call AI training. Far fewer have a workforce that has been through it, and fewer still have training built around the work people actually do.

The report also places organisations on a pathway, and the distribution is humbling: awareness 21%, exploration 19%, integration 7%, strategy 4%, scaling 1%. If you feel behind, you are in the biggest group. Over 44% of those organisations reported using AI tools daily while sitting at those early stages, which is exactly the mismatch that makes training urgent.

What you are actually buying

Corporate AI training is sold as a day, but a day is the delivery mechanism, not the product. Six things can be in the scope, and each one has a price. Knowing which you are paying for is most of the skill in buying it.

A quote covering only the second row is cheaper for a reason. Our guide to choosing an AI training provider goes through the questions that separate the two, and what corporate AI training covers defines the category if you are starting from scratch.

Six tests the evidence says to apply

The SKAI report distils its findings into six principles it calls PRIMES, with a checklist against each. They are written for providers, which makes them useful to buyers. Turn each into a question and ask it on the call.

• Practical. Does the training link to real tasks and decisions in the learner's role, and does it say plainly when AI should not be used?
• Reachable. Do people get paid, protected time to attend, and do the materials work for staff with very different levels of digital confidence?
• Integrated. Does it fit the systems and standards people already use? The report goes further than most and says baseline AI training should be mandatory before staff use workplace AI tools where organisational data, confidential information, regulated activity, safety or professional judgement are involved.
• Modular. Is it broken into short units people can enter at their own level? The report found 30 to 90 minute stackable modules are often more practical than longer courses, particularly for smaller organisations and frontline staff.
• Expandable. Do the skills transfer when the tool changes, and can the same core be rolled out to a second team without a redesign?
• Sustainable. Is there a plan to review the content after delivery, and does the provider monitor confidence, quality of use and decision making rather than attendance?

That last test is the one most often skipped. The report suggests revisiting training at three and six months so people can reflect on how their working habits actually changed, which is also roughly when the honeymoon wears off.

The failure modes that waste the budget

The same report lists thirteen pitfalls drawn from its workshops. Eight of them are visible in a proposal, before you have spent anything.

• The training is not linked to real work, so it is too general or too tool focused for anyone to apply.
• One approach is used for everybody, regardless of confidence, access or experience.
• Staff are not given the time, devices or connectivity the session assumes.
• People already using AI informally are treated as beginners, and switch off in the first hour.
• It moves too fast or too technically, which loses exactly the people who needed it most.
• It is built on one specific tool, so the capability expires when the tool does.
• Leadership does not prioritise it, link it to organisational goals or visibly use AI themselves.
• Impact is never measured, so the training gets deprioritised at the next budget round.

The barrier data is blunt about the practical side too. Employers named cost at 42% and limited availability at 37%. Both respond to tighter scope: a short, sharp session for the right ten people beats a broad rollout nobody finishes.

What businesses reported getting back

The survey asked what changed after training. Half of organisations, 50%, reported increased productivity and efficiency. A further 26% said training enabled new responsibilities or projects, and 12% reported improved confidence with digital tools.

Those categories are about capacity and scope rather than headcount, which is worth noting when you write the business case. The claim you can defend is that the same people take on more, not that you need fewer of them. If you want a way to evidence it, our method for measuring AI ROI starts from a baseline taken before the training rather than a survey taken after it.

The report is also clear that ongoing skills gaps are concentrated at the technical end, named by 67% of employers, against 32% for responsible AI and 10% for non technical skills. That ordering matters when you decide who gets trained in what.

What it costs, as far as anyone publishes

Most providers selling private corporate training publish nothing, because the number moves with headcount, format, duration and how much is built around your work. A few publish list prices for open courses, which at least gives you an anchor.

Learning Tree posts £730 per delegate for a one day introduction to AI. Elite Training Centre posts £795 + VAT per delegate for a one day Copilot and generative AI course. The Coders Guild posts £1,285 + VAT per person for a six session programme for leaders. All three were checked on 25 September 2026.

Per seat arithmetic stops making sense quickly once you are training a whole team, because you are paying repeatedly for content that does not change. Our breakdown of what corporate AI training costs works through the pricing models and where each one is the honest choice.

How to brief a provider in one page

Most weak quotes are the result of a weak brief. Six lines is enough to make quotes comparable.

• Who is being trained, and what those people do all day.
• The three recurring tasks you want faster or better, named specifically.
• The tools they already have licences for, and the ones they are not allowed to use.
• What data must never leave the building.
• The format, and how much protected time people are actually getting.
• How you will judge it in three months.

Send that and the proposals arrive on the same terms. Without it you are comparing day rates and hoping. If you want to see where your team starts before you write the brief, the AI proficiency assessment is free and takes about five minutes, and our UK AI training is scoped from exactly this kind of brief.