BCG surveyed 11,749 people for its 2026 AI at Work report. Seventy-two per cent said the skills expected of them had changed. Only 36% felt they'd had enough training to cope, a figure that hadn't moved since 2025 (BCG).
So most businesses are buying AI tools faster than they're teaching people to use them. And much of the AI training for employees that does happen is a one-off session nobody remembers by Friday.
The short answer: AI training for employees works when it's built around people's real tasks, uses the tools and data rules your business has actually approved, and keeps going after the first session. Teach everyone the basics and the risks, go deeper by role, train managers to redesign work, and measure changed behaviour rather than course completions. Generic prompt workshops on their own rarely change anything.
Key points:
- Training gaps are the norm: only 36% of workers in BCG's 2026 survey felt adequately trained.
- For providers and deployers within the EU AI Act's scope, Article 4 of the EU AI Act requires you to take measures to support AI literacy among staff who use AI.
- Free foundation courses exist in the UK. The real cost is staff time and the role-specific work only you can design.
- The tool is the cheap part. Changing how people work is the expensive part.
Why doesn't most AI training for employees work?
Because it's designed as an event, not a change in how work gets done.
The typical pattern goes like this. The company buys licences. Someone runs a webinar on prompting. Usage spikes for a fortnight, then settles back to the same handful of enthusiasts who were using it anyway. Everyone else goes back to the old way, because the old way still works and nobody told them what's allowed.
The research points at the same causes. Gallup found that only 15% of US employees said their organisation had communicated a clear plan for integrating AI, and almost half of workers who used AI said they'd been offered no training at all (Gallup). BCG's 2026 data adds a sharper detail: among frontline staff who already use AI regularly, 66% get limited or no guidance on what to do with the time it saves (BCG).
That last one matters. Saving an hour a day is pointless if the hour disappears into email. Training that ignores what happens next is training for a demo, not for the business.
What should AI training cover?
Good AI upskilling has four layers. Most programmes only deliver the first.
- Literacy. What these tools are, what they're good and bad at, why they make things up, and how to check their work.
- Your rules. Which tools are approved, what data can and can't go in, when a human must review output, and who to ask. This should come straight from your AI governance framework.
- Role-specific practice. The five or six tasks in each job where AI genuinely helps, practised on realistic examples from your own work.
- Redesigning work. For managers: which steps in a process change, what gets checked, and where the saved time goes.
Skip the theory of neural networks. Nobody in accounts payable needs it. What they need is to know that the tool will confidently invent a supplier's VAT number, and what to do about that.
Who needs what level of training?
Not everyone needs the same course. A practical split looks like this:
| Group | What they need | Rough time |
|---|---|---|
| Everyone | AI literacy basics, your approved tools, data rules, how to check output | A short foundation module, then refreshers |
| Regular users by role (sales, finance, HR, operations, support) | Hands-on practice on their own recurring tasks, with templates they keep using | A few short sessions over several weeks |
| Managers | How to redesign team workflows, set expectations, review AI-assisted work | Workshops tied to real processes |
| Leadership | Where AI creates value, risk and regulation, how to read results | A focused briefing, then regular reviews |
| Champions and builders | Building automations and agents, testing, handing over safely | Ongoing, project-based |
The managers' row is the one most programmes leave out. BCG found only 33% of frontline employees say leadership communicates clearly about AI (BCG). If managers don't change how they allocate and review work, staff learn quickly that the training was optional.
Is AI literacy training a legal requirement?
In the EU, effectively yes, though not in the form many training vendors suggest.
Article 4 of the EU AI Act has applied since 2 February 2025. It was amended in July 2026 by the Digital Omnibus on AI, Regulation (EU) 2026/1744. The amended text says providers and deployers of AI systems "shall take measures to support the development of AI literacy of their staff" and others using AI on their behalf, and it no longer requires them to guarantee any particular level for any individual (EUR-Lex).
The European Commission's guidance fills in the practical detail (European Commission):
- "Deployer" includes ordinary businesses using AI tools at work, not just AI developers.
- No particular certificate is prescribed. Keep records of proportionate measures and review whether they address the roles and risks involved.
- There's no mandated format, but relying only on a tool's instructions for use is generally not considered enough.
- It reaches organisations outside the EU when the AI system is used in the EU or affects people there.
- National authorities take over supervision and enforcement from August 2026.
For UK businesses, that last point is the one to check. If you have EU customers, staff, operations or outputs used there, assess the Article 2 scope rather than relying on a customer-address shortcut. The full picture is in EU AI Act compliance for UK businesses.
So be wary of anyone selling an expensive "AI Act certificate". The law asks for reasonable effort and a paper trail, not a badge.
How much does AI training for staff cost?
It varies too much for a single honest number, so think about it in three parts.
Foundation literacy can be free. In January 2026 the UK government expanded its AI Skills Boost programme, making benchmarked foundation courses free for every UK adult, some taking under 20 minutes, with a target of 10 million workers trained by 2030 (GOV.UK). The same announcement cited research finding only 21% of UK workers feel confident using AI at work.
Role-specific training is where money goes. Off-the-shelf courses from vendors and platforms are priced per seat. Custom workshops built around your own processes are priced per day by trainers or consultants. Ask for quotes on the second, and judge them on how much of the content is about your work rather than theirs.
Staff time is the biggest line. It's also the one nobody budgets. A programme that takes 50 people out of their jobs for a few hours each, several times, is a real cost. Plan for it, and pick the quieter weeks.
The cheapest mistake to avoid is paying for premium licences that nobody has been shown how to use. Train before or alongside rollout, not six months after. If you're still choosing tools, ChatGPT vs Copilot vs Claude for business covers the trade-offs.
How do you know if it is working?
Course completions tell you people clicked through. They don't tell you anything changed. Better measures:
- Active use of approved tools, by team, over months rather than the launch week.
- Time on target tasks, measured before and after for the handful of tasks you trained on.
- Quality and error rates on AI-assisted work, especially anything customer-facing.
- Policy incidents, such as sensitive data pasted into unapproved tools. Interpret reports carefully: an initial rise can mean staff are better at spotting and reporting problems.
- Where saved time went. If nobody can say, the gain probably isn't real yet.
Pick a baseline before you start. How to measure AI ROI covers the method in more detail.
My take
I use AI every day in my own work: research, synthesis, first drafts and repeatable workflows. I also run Ascendancy IQ, a lab where AI agents do real work under human direction, with approval gates before anything consequential happens. The most useful skill that has taught me isn't prompting. It's supervision: knowing what to delegate, how to check it and when to say no.
That's the skill most AI training for employees misses. As tools take on more steps on their own, the job shifts from doing the task to directing and reviewing it. BCG's respondents noticed the same thing, with nearly half saying their roles had shifted towards managing and directing AI.
A common mistake is to hand the whole thing to HR or L&D as a content problem. It's an operations problem. The person who owns the process should own the training for it.
And don't wait for perfect. Start with one team, one workflow and a clear rule sheet. If you haven't checked whether your data and processes are ready for that, the AI readiness checklist is a sensible first stop.
How to run AI training for employees
- Find out what's already happening. A short anonymous survey on which tools people use and for what. You'll find unofficial use. Treat it as information, not a disciplinary matter.
- Set the rules first. Approved tools, data do's and don'ts, review requirements. One page is fine.
- Run a foundation module for everyone. Use free courses where they fit, plus a session on your own rules.
- Pick two or three high-volume workflows. Build role-specific practice around them, using realistic examples with sensitive data removed.
- Train the managers of those teams. Agree how the work changes and where saved time goes.
- Appoint champions. One person per team who gets deeper training and becomes the first port of call.
- Keep a record. Who was trained, on what, when. That's your Article 4 evidence if you need it.
- Measure, then expand. Review use and results after 60 to 90 days, fix what didn't land, then move to the next workflows.
This fits inside a wider rollout plan. See how to implement AI in your business for the rest.
Test supervision, not prompt recall
Give learners an output with a plausible invented fact, an inappropriate data request and a case that needs human escalation. Ask them to identify the problem and show the approved next step. Then repeat the exercise on their own workflow after a few weeks. That tells you more than whether they can remember the trainer's favourite prompt.
Frequently asked questions
What is AI literacy training for employees? It's training that helps staff understand what AI tools do, where they fail and how to use them safely in their job. It should also cover your organisation's own rules on approved tools and data.
Is there free AI training for employees in the UK? Yes. The government's AI Skills Boost programme offers free foundation courses to every UK adult through the AI Skills Hub. They're a good baseline, but you'll still need training on your own tools, rules and workflows.
Do UK companies have to provide AI training? There's no UK equivalent of the EU's AI literacy duty. UK providers and deployers within the EU AI Act's scope must address Article 4. Establish that scope for each use rather than assuming every interaction with an EU person qualifies.
How long should AI training for employees take? A foundation module can be under an hour. Role-specific training works best as short sessions spread over several weeks, with practice on real tasks in between, rather than one long day.
What's the difference between AI upskilling and AI literacy? AI literacy is the baseline everyone needs: what the tools are, their limits and the rules. AI upskilling goes further, teaching specific roles to use AI to do their work differently.
Sources
- AI at Work 2026: Strategy Matters More Than Tools, Boston Consulting Group, June 2026
- AI in the Workplace: Answering 3 Big Questions, Gallup, October 2024
- Regulation (EU) 2026/1744 (Digital Omnibus on AI), EUR-Lex, July 2026
- AI Literacy: Questions and Answers, European Commission
- Free AI training for all as government and industry programme expands, GOV.UK, January 2026
If you want an AI training programme built around how your business actually works, rather than a generic course, let's talk.
