"How do I train my employees to use AI?" Not with a big launch day and a long slide deck. People learn a tool by using it on the work in front of them, with a few clear rules and someone nearby to ask. This article is for the owner of a company with ten to two hundred people who has noticed that some staff are already using AI, some are nervous, and most are waiting to be told what is allowed. It gives a plain order of work: who to start with, what to teach first, how long each session should be, how to tell whether it is helping, and what to leave alone for now.
I was COO and then President of a public company that closed more than forty acquisitions, and I ran the integration on most of them, so I have seen how a company's habits show up when something new lands on a team and nobody has said how it should be used. I am a lawyer by training. Now I write fixed-fee AI plans for CEOs of private-equity-backed and owner-led mid-market companies through 013 Labs. This is not legal advice. Where the law could apply to your company, talk to your own counsel.
Start with the people already using tools
You probably have users already. In most companies somebody is quietly using a free tool to tidy emails, summarise a long document or draft a quote, and has not mentioned it because nobody asked. Do not start by lecturing the team. Start by asking, in a friendly way, who is using what and what for. Make it safe to answer: the aim is to learn, not to catch anyone out.
Those early users are your best teachers, because they already know which tasks the tool helps with and where it lets them down. One of them is often your internal champion, the person others turn to with questions. Spot your AI champion describes what that person looks like and how to find them. Who owns AI in a company covers what to give them in writing. Give them a little time in the week and some backing from you, and let them lead the first sessions. A colleague who says "here is how I do it on our own orders" is far more convincing than an outside speaker with general advice.
Train on your own tasks, not on features
Generic training teaches what the tool can do. What your people need is how it helps with the job they have on Monday. So build every session around a task the company really performs: answering a type of customer email, writing a job description, summarising a supplier's terms, drafting a weekly report, tidying meeting notes.
Ask each team for two or three tasks that eat their time and are done the same way again and again. Those are the training material. A person who sees the tool do a piece of their own work, and then checks and improves it, learns more in an hour than from a long tour of menus. If you are unsure which task to start with, picking your first AI project is a good way to choose, and can AI answer customer emails for a small business is a worked example of one common starting task.
Teach the data rule and the checking habit first
Two things come before any clever technique, and both fit in a few minutes.
The data rule. One sentence about what may and may not go into an outside tool. Something in this spirit: "Nothing about a customer, an employee or a supplier, and nothing we have promised to keep private, goes into any tool that is not on the approved list, on a company account." Staff should be told which tools are approved, that they must use the company account and not a personal one, and that bank details, card numbers, identity documents and health information stay out entirely. Is ChatGPT safe for business use explains what a business plan changes about your data and what it does not. If you have nothing written, do I need an AI policy for my small business shows how to put the rule on one page.
The checking habit. A tool writes fluent text, and fluent text can be wrong. Teach everyone that the person who sends the work owns it. Names, figures, dates and anything that sounds like a promise get checked against the source before the work goes out. A good habit to teach: ask yourself what in this draft would hurt if it were wrong, and check that part first. Make it a rule of the house that "the tool said so" is never an answer.
Staff who learn these two things first make fewer mistakes later, and they are easier to trust with more.
Short sessions on a real task
Keep each session to about an hour, and make most of it practice. A workable shape:
- A few minutes to restate the data rule and the checking habit.
- A short demonstration by the champion, on a real task from that team.
- The bulk of the hour with each person trying the task on their own work, with the champion walking around.
- A few minutes at the end for each person to say what worked, what did not and what they will try next week.
Then stop. A long session ends with people tired and remembering little. Several short ones, a week or two apart, give people time to try the tool on live work and come back with real questions. Run it with one team first, learn what to change, and only then widen to the next.
Pair people up and have a place for questions
People learn faster beside a colleague than alone. Pair a confident user with a hesitant one, ideally from the same team so they share the same tasks, and let them compare how they each did a task. The more confident one learns by explaining; the other gets a patient answer without feeling foolish in front of the room.
Give questions a home. A shared channel or a simple shared document works, where anyone can ask "has anyone used it for this?" and the champion or a colleague answers. Keep a running list of what people have found useful and what has gone wrong. That list becomes your best training material, because it is written in your company's own words about your own work. If nobody on staff is technical, AI for a small business with no IT department covers how to run all this without one. If you are weighing a new hire instead, hire for AI or train your current staff sets out when each makes sense.
Measure from your own records
Do not ask people whether they feel more productive and stop there. Use what the company already keeps. Pick a task before you start and note how long it took, how many rounds of correction it needed, or how many came back with a mistake. Your ticket system, your order records, your shared drive and your weekly reports probably hold this already. Look at the same measures after a few weeks of use.
Add a simple log: the task, whether the tool was used, what the person changed, and whether it saved time. A shared spreadsheet is enough. Be honest about the result. If a task is no faster, or the checking takes as long as doing it by hand, say so and move on to a different task. The aim is to learn where it helps, not to prove that it does. For the cost side, how much does AI cost for a small business breaks the spend into plain parts.
What not to train on yet
Set the boundaries as clearly as the permissions. For now, leave these out of training, and keep them with a named person:
- Anything that sets a price, makes a delivery promise or grants a refund.
- Hiring, firing, pay and performance decisions about individuals.
- Legal, tax and contract wording, which belongs with your own advisers.
- Any task that needs customer, employee or supplier details you have not cleared under the data rule.
- Letting a tool send anything to a customer or supplier without a person reading it first.
You can revisit the list once the team has shown good judgment on easier work. It is far easier to widen a rule than to repair a mistake that has already gone out.
What changes for managers
Managers carry most of the change, and they are often left out of the plan. Three things shift.
They review differently. Work now arrives faster and may look polished while hiding an error. Managers need to ask how it was checked, not only whether it is finished.
They set the tone. If a manager uses the tool openly and talks about what went wrong, staff will do the same. If a manager treats it as a secret or a threat, people will hide their use, and hidden use is the risky kind.
They protect the time. Learning takes time away from the day's work at first. A manager who expects the same output from the same hours during the learning weeks will get people who skip the practice. Give the champion and the team some room, and say so out loud.
Brief your managers before the team sessions, so that they are not the last to find out and are able to answer the first questions.
Some people will not take to it
Expect it, and let it be fine. In any team some people will take to the tool at once, some will use it a little, and some will not want it. A person who does careful, skilled work without it is still doing good work. Do not make use a test of loyalty, and do not post a league table of who uses it most.
What you can ask of everyone is smaller: that they follow the data rule, that they check anything that goes out under their name, and that they tell you honestly what helps and what does not. Those who want to go further will, and they will pull others along. Pressure tends to produce hidden use and resentment, and neither helps the company.
Picture it going wrong
Picture a distribution company of ninety people. The owner books a single half-day session for everyone, with a general talk on what AI can do and no written rule on data. The talk is exciting. Over the next weeks a few people start pasting customer details into personal accounts, a manager sends a supplier a figure that a tool invented and nobody checked, and the quieter staff conclude it is not for them and say nothing. Nobody was careless on purpose. Three months later the owner has no idea who is using what, no record of whether it helped, and some uneasy staff. The repair is modest: find the people already using it and the champion, write the one-sentence data rule, teach the checking habit, run short practice sessions on real tasks one team at a time, pair people up, keep a log, and make it clear that no one is judged for choosing not to use it. This is a hypothesis, a composite and not a client.
Mistakes that repeat
One big launch session and nothing after. Teaching features instead of the company's own tasks. No written data rule. Skipping the checking habit. Leaving managers out of the plan. Measuring by opinion alone. Treating non-users as a problem. Widening the tool's permissions before the team has shown good judgment. Skipping your own counsel where the law could apply.
Frequently asked questions
How do I train my employees to use AI?
Start with the people already using it and your internal champion, train on your company's own tasks, and teach the data rule and the checking habit first. Run short sessions of about an hour with practice on a real task, pair people up, and keep a place for questions.
How long should AI training take?
Keep each session to about an hour, and run several a week or two apart rather than one long day. Start with one team, learn what to change, then widen to the next.
Do I need an outside trainer?
Not necessarily. A champion from your own staff, teaching on your own tasks, is often more convincing than a general talk. An outside adviser can help set the rules and the plan, but the practice happens on your own work.
What should staff never put into an AI tool?
Bank details, card numbers, identity documents and health information, plus anything you have promised to keep private. Set a one-sentence data rule, use company accounts, and keep personal accounts out of it.
How do I know the training worked?
Use your own records. Note how long a task took, how many corrections it needed or how many came back with a mistake before you started, then compare after a few weeks, and keep a simple log of what people changed.
What if some employees refuse to use AI?
Let them. Ask only that everyone follows the data rule, checks what goes out under their name and tells you honestly what helps. Pressure tends to produce hidden use, which is the risky kind.
Is there anything the law requires me to teach?
I cannot give legal advice, and it depends on your location, your industry and your contracts. Ask your own counsel what applies to your company before you write the rules.
Not sure where you stand? Take the 3-minute AI readiness scorecard.