How to train your employees on AI tools

Why most AI training doesn't stick

A common pattern: the firm buys seats, someone runs a general demo, and three months later only a handful of people use the tool. The rest tried it once, got a mediocre answer and went back to the old way.

The problem is rarely the tool. People were shown what AI can do in general rather than how to do their own Tuesday-morning task faster. Effective training starts with each role's real work and gives people a reason to come back.

A second problem is fear. Some staff worry that getting good at AI means writing themselves out of a job, and some worry about getting in trouble for using it wrong. Address both directly in the kickoff: explain what the firm wants to use the time for, and make clear that asking questions and reporting mistakes is expected.

Before training: three things to have in place

Training goes better when the basics are settled first. Otherwise every session turns into a debate about which tool is allowed.

If any of these are missing, delay the training by a week or two rather than training on tools people cannot use yet.

• Approved tools on business accounts, already set up with company logins.
• A one-page AI policy covering data tiers, review rules and how to request new tools.
• Two or three real examples per role, such as a recent proposal, a messy call note or a long client email thread, with sensitive details removed.

A four-week training plan

This plan fits around normal work in a 10 to 100 person firm. Sessions are short and practical.

Schedule sessions at quiet times in your business cycle where possible. A tax practice will get little attention in March, and a landscaping or roofing company will struggle in peak season. A short session in a slower month beats a long one when everyone is distracted.

• Week 1, kickoff (45 minutes, everyone): what the approved tools are, the three data tiers with your own examples, the review rule, and a live demo on a real task from your business.
• Week 2, role sessions (30 to 45 minutes each): separate short sessions for front office, operations, sales and professional staff. Each person completes two of their own tasks with the tool during the session.
• Week 3, practice and prompt library: staff save prompts that worked into a shared document or Teams or Google Chat space. The champion reviews and tidies them.
• Week 4, check-in (20 minutes per team): what is working, what is not, and what people still avoid. Update the policy or library based on the answers.

What to teach each role

Keep examples specific to the work. A few that fit common service-business roles:

Keep each role session small, ideally under eight people, so everyone gets hands-on time. Recording the kickoff is useful for new hires, but role sessions work better live, because the questions people ask are the training.

• Office manager or front desk: summarizing long email threads, drafting polite payment reminders, turning voicemail transcripts into tidy call logs.
• Dispatcher or scheduler: drafting customer arrival messages and rescheduling notices from a short template.
• Estimator or project manager: turning site notes into a scope outline, checking a scope for missing items against a checklist.
• Paralegal, bookkeeper or associate: summarizing documents for a professional to review, drafting first-pass client letters, extracting dates and amounts into a table.
• Partners and owners: preparing for client meetings from past notes, drafting proposals from a call summary, reviewing others' AI-assisted work.

Teaching review as a skill

The most important habit is checking the output. Teach it explicitly. Show an example where the AI tool got a figure, a date or a name wrong, and walk through how you would catch it.

Give people a short review checklist: Are the facts and numbers traceable to the source? Is anything stated that we did not tell it? Does the tone fit the client? Would I sign this? For regulated work, such as legal, tax or clinical content, make clear that the responsible professional must verify against primary sources every time.

Ask each team lead to choose the two tasks for their session. They know which chores people dislike most, and those make the most persuasive demonstrations.

Sample exercises for the first session

Live exercises on your own material teach more than slides. These work well in a first session. Prepare de-identified versions of real documents in advance.

Make it easy to ask for help after training ends. A dedicated chat channel where people post a task they are stuck on, and the champion or a colleague suggests an approach, keeps learning going without more sessions. Encourage people to share failures as well as wins, since those teach the review habit.

• Paste a long, messy email thread and ask for a five-line summary with open questions and who owns each one. Then compare it to the thread and find anything it missed.
• Give it rough call notes and ask for a follow-up email in your firm's tone. Ask for a shorter version, then a more formal one, so people see that iteration is normal.
• Ask it to turn a procedure from your office manual into a checklist, then have the person who owns the procedure mark what is wrong.
• Ask it a question it cannot know, such as your current price for a service, and show how confidently it may answer. This is the best way to teach review.
• Give it a scope of work and a checklist of required items, and ask what is missing. Discuss which flags were useful and which were noise.

Champions, follow-up and measurement

Pick one or two people who already use AI well and give them an hour a week to help colleagues and curate the prompt library. Peer help usually beats formal training for adoption.

After 30 days, look at a few simple measures: how many people use the approved tool weekly, which tasks they use it for, and whether reviewers are catching more or fewer errors. Ask people which tasks they still do the old way, and why.

Repeat a short refresher when tools change or new staff join, and include the AI policy in onboarding. If you would like help designing role-based sessions around your own workflows, InTheory includes team training in its Embedded plan.