The short answer
Train your team on one tool first, tie every session to three real tasks they already do each week, and give them a one-page policy so they know what is safe to paste in. Most people reach a useful working level in a few weeks, not months. The mistake is buying licences and hoping people work it out on their own.
Your staff are already using AI. That is not a prediction, it is McKinsey's number: the share of employees using AI at work went from 30 percent in 2023 to 76 percent in 2025. Some of your team opened ChatGPT this morning. The question is no longer whether they use it. It is whether they use it well, and whether they use it safely.
Untrained use is where the risk sits. Confidential client data pasted into a free tool. A confident, wrong answer sent out in an email. An afternoon lost to prompts that go nowhere. Training is how you turn quiet, ungoverned use into a genuine advantage. Here is how to do it without losing a year to the project.
Why does most AI training fail?
Because it starts with the tool instead of the task. The classic version is a one-hour "Introduction to ChatGPT" webinar. Forty people watch a screen share, nod along, go back to their desks, and nothing changes by Friday. The opposite failure is just as common: buy the Copilot licences, send the welcome email, and assume adoption will look after itself.
There is a governance gap underneath all of it. The CIPD found only 31 percent of UK employers have a formal AI policy, even though staff use the tools every day. So the small amount of training that does happen is usually ungoverned and unmeasured. People are experimenting in the dark, and nobody can see what is working.
What does good AI training actually look like?
Good training is boring in the best way. It is specific, it is repeated, and it is tied to work people already do:-
The four things that separate training that sticks
- Grounded in real tasks. People learn on the actual brief, the actual spreadsheet, the actual customer email, not a toy example.
- Role-specific. What a marketer needs from AI is not what finance needs. Generic training gives everyone something that fits nobody.
- Short and repeated. A half day to start, then light practice and follow-up, beats one long course everyone forgets.
- Safe by design. A one-page policy sits alongside the training so people know exactly what they can and cannot put into a tool.
A practical five-step plan
-
Start with one tool
Pick one general assistant and get properly good at it before adding anything else. For most teams that is ChatGPT or Claude, or Copilot if you live in Microsoft 365. Spreading a team thin across five tools on day one is like buying a Ferrari and only using it to drive to Lidl. Depth beats breadth.
-
Pick three real tasks per team
Sit with each team and name three things they do every week that are slow, repetitive or a pain. First-draft copy, meeting notes, research summaries, replying to routine enquiries. Those three tasks become the whole training. Everything is taught on them.
-
Write a one-page AI policy first
Before anyone is trained, agree the rules. Which tools are approved, what data must never be pasted in, and who to ask when unsure. One page, plain English. This is the single cheapest piece of risk management you will do all year.
-
Train in short, hands-on sessions by role
Half a day, jargon free, everyone on their own laptop working on the real tasks from step two. People should leave with prompts saved and outputs they can reuse the same afternoon. If they walk out with only notes, it did not work.
-
Measure the hour saved, then widen
Pick one number per team. Hours saved on a weekly report. Time to first draft. Enquiries answered per day. Check it a month later. When you can point at a real saving, the rollout to the rest of the business sells itself.
Where should you start by team?
You do not have to train everyone at once, and you should not try to. Start where the win is obvious. This is a rough map of where each function tends to get value first.
| Team | Best first use | What to train on | Quick win |
|---|---|---|---|
| Marketing | First-draft copy and content repurposing | Briefs, tone of voice, turning one asset into five | Halve the time to a usable first draft |
| Sales | Research and proposal drafting | Prospect research, tailoring pitches, follow-ups | More time in front of clients, less at the keyboard |
| Operations and admin | Summarising and drafting | Meeting notes, documents, routine replies | Clear the low-value inbox in half the time |
| Finance | Explaining and checking | Plain-English summaries, spotting anomalies, formulas | Faster month-end narrative |
| Leadership | Thinking and deciding | Scenario testing, board prep, pressure-testing plans | Better questions in the room, sooner |
How long does it take?
Less time than people expect. Built around real tasks, most people reach a useful working level in a few weeks. That is the whole secret. Confidence across a full team builds over a quarter with light, repeated practice. The teams that never get there are almost always the ones who booked one big session, ticked the box, and moved on.
The mistake that costs the most
Buying licences and skipping the habit. The subscription is the easy part. The value only shows up when using AI becomes a reflex for a specific job, and that only happens with practice on real work and a bit of follow-up. A tool nobody has built a habit around is just a line on the invoice. Spend less on licences and more on making three tasks per team genuinely faster. That is where the return is.
How to tell whether the training worked
Most AI training is never measured, which is why so much of it can fail without anyone noticing. You do not need a framework — you need three numbers and a date in the diary six weeks out.
Adoption, honestly counted. How many people used an approved tool on a real work task this week? Not licences issued, not attendance, actual use. This is the number that exposes training that felt good in the room and changed nothing on Monday.
Time on named tasks. Pick two or three specific recurring jobs before you train — the weekly report, the standard quotation, first-draft client emails — and ask people roughly how long each takes. Ask again at six weeks. Rough self-reported figures are fine; you are looking for a direction, not an audit.
Quality and rework. The number nobody wants to collect and the one that keeps you honest. Are drafts needing more correction than before? If speed has gone up and quality has gone down, you have taught the tool and not the checking, and that is fixable.
Then ask one open question: what did you stop using it for, and why? The answers are usually the most useful thing you will learn, because they point at where the training was thin rather than where it was popular.
Skills decay, and what to do about it
Six weeks after a good session, a predictable thing happens. Two or three people have got noticeably better, most have settled into using it for one narrow task, and a few have quietly stopped. This is normal, it is not a failure of the training, and it is entirely addressable — but only if somebody expects it.
The cause is partly human and partly that the tools keep moving. A feature that did not exist when you trained now solves the problem someone gave up on. Nobody finds out, because finding out is not anyone's job.
Three low-effort habits hold the gains. Keep a shared place for prompts that worked, so good practice spreads without a meeting. Spend ten minutes in an existing team meeting each month on one thing somebody found useful — not a presentation, just a demonstration on real work. And schedule one short refresher a few months after the main session, built around the problems people actually hit rather than a repeat of the curriculum.
The refresher is where the real gains sit. In the first session people learn what the tools do; by the second they know what they need, and the questions get much better.
Training the people who do not want to
Every team has some, and how you handle them decides whether adoption reaches two thirds of the business or a third. Resistance is usually one of four things, and they need different answers.
"This is going to replace me." The most common and the least often said aloud. It needs answering directly and honestly by someone with the authority to mean it. If the intention is to redeploy time rather than reduce headcount, say so plainly; if you cannot say that, do not pretend otherwise, because people will work out which it is regardless.
"I tried it and it was rubbish." Almost always someone who asked a vague question, got a generic answer and concluded the technology was overrated. This one is easy: sit with them on a real task, show the difference context makes, and they often become the most enthusiastic person in the room.
"It is not accurate enough for my work." Frequently a fair point, from someone whose work genuinely demands precision. Agree with them, then move the conversation to the parts of their week that are not precision-critical — the first draft, the summary, the reformatting. Respecting the objection earns you the rest.
"I do not have time to learn something new." Usually true, and usually the person with the most to gain. The answer is one task, small, that saves them something this week. Nothing persuades like an afternoon back.
One thing to avoid: making enthusiasm compulsory. Mandating attendance is reasonable; mandating delight produces people who nod through the session and change nothing. Let the results do the arguing.
How to Train Your Staff to Use AI – frequently asked questions
How do I train my staff to use AI?
Start with one tool, tie every session to three real tasks each team already does, and put a one-page policy in place first so people know what is safe to paste in. Train in short, hands-on sessions by role rather than one big webinar, then measure the time saved and widen from there.
How long does AI training take?
Most people reach a useful working level in a few weeks, not months, once training is built around tasks they actually do. Confidence across a team builds over a quarter with light, repeated practice rather than a single one-off course.
Do we need an AI policy before training?
Yes. A one-page policy that says which tools are approved and what must never be pasted into them should come before or alongside training. Only 31 percent of UK employers have a formal AI policy while staff use the tools daily, so most training happens ungoverned. Our AI Policy Workshop covers exactly this.
What is the best way to train a marketing team on AI?
Train on the work: campaign briefs, first-draft copy, research summaries and repurposing content across channels. Role-specific, hands-on sessions on real projects beat a generic introduction, because the team leaves with prompts and outputs they can use the same afternoon. See AI for Marketing Teams.
Should we train everyone at once or start with one team?
Start with one team that has a clear, repetitive task and an appetite for it. A visible early win gives you internal proof and champions, which makes the rollout to the rest of the business far easier than trying to move everyone on day one.
How much does AI training cost for a UK business?
It varies by format and team size. The Oxford AI School runs half-day team sessions and bespoke programmes, with most team workshops coming in under 5,000 pounds. The bigger cost is usually the time lost to untrained, unfocused use, not the training itself. Get in touch for a quote.
How do we measure whether AI training worked?
Track three things and check at six weeks: how many people used an approved tool on a real task this week, roughly how long two or three named recurring jobs now take compared with before, and whether rework has gone up. Then ask what people stopped using it for and why — that answer usually shows where the training was thin.
Why do staff stop using AI a few weeks after training?
Because skills decay and tools move. People settle into one narrow use, hit a limitation, and quietly give up — often for something a newer feature would now solve. Counter it with a shared place for prompts that worked, ten minutes in an existing monthly meeting on one useful find, and one short refresher a few months later built around the problems people actually hit.
How do you handle staff who do not want to use AI?
Work out which objection it is, because they need different answers. Fear of replacement needs a direct, honest response from someone with authority. 'I tried it and it was rubbish' usually means a vague prompt, and is fixed by sitting with them on real work. Accuracy concerns are often fair — agree, then focus on the non-critical parts of their week. 'No time' is best answered with one small task that saves them an afternoon.
Which Oxford AI School training fits this need?
Choose practical training around your team, tools and real work. Sessions run in Oxfordshire, London, across the UK and internationally through live online delivery.
Want this done properly for your team?
The Oxford AI School runs half-day, jargon-free sessions and bespoke programmes for UK teams. We help you choose the right one or two tools and get properly good at them, rather than collecting subscriptions.
Build a plan for your team See our programmes
Harry Lang runs The Oxford AI School, teaching teams and businesses the basics of AI in a fun, easy and practical way. A marketing leader turned AI trainer, he can be reached at Learn@TheOxfordAISchool.com.