For everyday office use, plan six to twelve hours per person over the first two months: a two-hour session on basics and rules, a few hours practising on real tasks, and short follow-ups. Cash cost can be close to zero using free vendor courses; the real cost is paid time, about $180 to $360 per person at $30 an hour.
Those are planning figures built from what a sensible programme contains, not an industry average. Field staff who only use a phone app need far less; the person who'll champion AI inside the business needs far more. The sections that follow give hours by role, the cash options with current prices, a costed plan for an eleven-person decorating firm, three ways to deliver the training, and the measures that show whether the time paid off.
Hours by role in the first two months
Not everyone needs the same training, and budgeting as if they do is the most common way to overspend. Split people by how much they'll use AI:
| Role | What they need to learn | First two months | Ongoing |
|---|---|---|---|
| Occasional user (for example, field staff with a phone app) | The one or two tasks they'll use it for, plus the data rules | 1 to 2 hours | Nothing regular |
| Everyday office user | Basics, your rules, three of their own tasks practised to a standard | 6 to 12 hours | About an hour a month of clinics or refreshers |
| Heavy or specialist user (estimator, marketing, bookkeeping) | The above plus more complex tasks, checking methods, working with files and data | 12 to 20 hours | 1 to 2 hours a month |
| AI champion | Everything above, plus how to help others, maintain prompts and spot misuse | 20 to 30 hours | 2 to 4 hours a month |
| Owner or manager | Rules, what to check, how to review use, plus their own practice | 4 to 6 hours, plus own practice | An hour a month reviewing |
For an everyday office user, the six to twelve hours break down roughly like this: a two-hour introductory session; an hour on your AI rules and what data may go where; three to six hours practising on their own tasks, spread over three or four weeks; and three or four follow-up clinics of 20 to 30 minutes. The practice is the part people cut and the part that decides whether anything changes. How to train staff to use AI covers the content of each stage; this tutorial is about sizing it.
Here's that breakdown filled in for one person: the sales coordinator at an illustrative events caterer, on a loaded cost of $27 an hour.
TRAINING PLAN: sales coordinator Champion: operations manager
Her three tasks: (1) enquiry email -> draft quote from the price list
(2) menu descriptions for proposals
(3) summary of tasting-session feedback for the chef
Week 1 Intro session (2 h) + AI rules and data (1 h) 3.0 h
Week 2 Practise task 1 on 6 real enquiries, checked by manager 1.5 h
Week 3 Practise tasks 2 and 3 on this week's real work 1.5 h
Week 4 Clinic (30 min); add her working prompts to the library 1.0 h
Weeks 5-8 Three 20-minute clinics; one 30-minute review 1.5 h
Total 8.5 h
Cost: 8.5 h x $27 = about $230
Free vendor course: taken in work time, inside week 1's intro hoursThe last line is a decision worth writing down. If the free course happens "in her own time", it quietly doesn't happen, or it happens resentfully. Count it in work hours and it's part of the $230.
What you'd pay in cash
The major AI vendors now give away solid introductory training, so the cash line can be very small:
- Claude Academy from Anthropic is free. Its catalogue includes non-technical courses such as AI Fluency: Framework and foundations (listed at about 4 hours) and AI capabilities and limitations (about 3.5 hours). See the Claude Academy catalogue.
- OpenAI Academy courses are free to anyone with a ChatGPT account, and include a track on applying AI at work.
- Microsoft Learn has free Copilot modules and learning paths for business users, useful if you're on Microsoft 365.
- Google AI Essentials on Coursera is a paid, self-paced option described as taking under 10 hours. It's sold through a Coursera subscription, listed at $49 a month after a seven-day free trial on the course page when checked in September 2026; Coursera's prices vary by country.
- An external trainer charges by the session or the day, and rates vary widely. The value is that the session can use your tools and your tasks. Get two quotes for the same brief; choosing an AI training provider for your team has the questions to ask.
One limit applies to every generic course: it teaches how AI works, not how your business uses it. None of them know your rules, your templates or which client documents must never be uploaded. Budget the generic course as a primer and spend the bulk of the hours on your own work.
The cost nobody invoices
Staff time is the main cost, and it's easy to underestimate because it doesn't arrive as a bill. Three pieces of it are worth naming:
- Training hours at full cost. Use each person's loaded hourly cost (wage plus employer on-costs), not their wage alone.
- The slower weeks. People work more slowly while they're learning a new way to do a familiar job. Don't schedule the practice period in your busiest month, and don't judge the tool in its first fortnight. The dip is small but real, and easy to put a number on: if two estimators each spend 10 hours a week on quotes and work about 20% slower for the first two weeks, that's 2 × 2 × 2 = 8 hours, about $256 at $32 an hour, which appears on no invoice.
- The champion's other work. Twenty-plus hours from your most capable person has to come from somewhere. Decide in advance what they'll stop or delay, or the training will quietly happen in evenings and then stop. Choosing and supporting an AI champion covers how to protect that time.
Worked example: training an eleven-person decorating firm
Take an illustrative painting and decorating firm: the owner, an office manager, two estimators, an administrator and six decorators. The firm has just rolled out an AI assistant for the office and a phone app for decorators to dictate job notes and snag lists. The office manager is the champion.
| Who | Hours each | Loaded hourly cost | Cost |
|---|---|---|---|
| Owner | 5 | $40 | $200 |
| Office manager (champion) | 24 | $30 | $720 |
| 2 estimators | 14 | $32 | $896 |
| Administrator | 8 | $26 | $208 |
| 6 decorators | 1.5 | $28 | $252 |
| Staff time, 74 hours in total | $2,276 | ||
| Cash: free vendor courses, plus one month of Coursera for the champion | $49 | ||
| Total | About $2,325 | ||
The estimators' hours are higher because their tasks are harder: turning site measurements and photos into itemised quotes and specification notes, and checking that AI hasn't invented quantities. The decorators' 90 minutes covered the phone app and one rule: no photos of customers' personal paperwork or family pictures in job notes.
A good share of the estimators' extra hours went on one exercise: checking AI quantity calculations where the office manager had seeded mistakes. One of the five practice sheets read like this (illustrative):
Room: 4 m x 3.5 m, ceiling height 2.4 m. Two coats on the walls.
Paint covers about 12 square metres per litre.
AI ANSWER
Wall area: (4 + 3.5) x 2.4 = 18 square metres
Two coats: 36 square metres
Paint needed: 3 litresThe arithmetic on each line is correct, and the answer is half what the job needs, because the perimeter is 2 × (4 + 3.5) = 15 metres, not 7.5. The real wall area is 36 square metres before subtracting the door and window, and two coats need about 6 litres. One estimator caught it in seconds; the other signed it off, which is exactly what the exercise is for. By the fifth sheet both were checking the formula behind each line rather than the sums on it. That habit is worth more than the time saved on drafting, and it's why heavy users need the upper end of the hours range.
The decorators' session was short enough to plan to the minute:
| Minutes | What happened |
|---|---|
| 0 to 10 | Why the firm is using the app, and the one rule about photos and paperwork |
| 10 to 40 | Each decorator dictates a snag list for a real job from last week while standing in the yard, as they would on site |
| 40 to 60 | Read the app's list back against what they said; fix the two most common problems (rooms named inconsistently, and paint colour names misheard) |
| 60 to 80 | Dictate a second list and send it to the office |
| 80 to 90 | Questions; who to ring when the app misbehaves |
Seventy of the 90 minutes are spent on their own work, and none on how AI works in general. For occasional users that ratio is right.
Was it worth it? By week eight the estimators were saving around three hours a week each on quotes, the office manager about three on correspondence, and the administrator about two on scheduling emails: eleven hours a week, worth roughly $330 at the office's average loaded cost of $30. At that rate the training cost is recovered in about seven weeks. Had the firm trained all eleven people for twelve hours each, the cost would have been about 70% higher for little extra saving, because the decorators don't need it.
What pushes the hours up
The ranges above assume one main AI tool, staff who are comfortable with office software, and mostly internal data. Each of these adds time, and it's better to budget for it than to discover it in week three:
- Sensitive client data. If people handle contracts, medical details or financial records, add an hour for the data rules and practice on what to anonymise before pasting. Getting this wrong costs far more than the hour. In one illustrative bookkeeping practice that cut the data hour to save time, a trainee pasted a client's full bank statement, account numbers and all, into his personal chat account to categorise transactions. It came to light only because he mentioned it proudly at a clinic. Undoing it meant a conversation with the client and the practice's data-protection adviser, which took far longer than the hour it had saved.
- More than one tool. Each extra tool, such as a meeting note-taker alongside the assistant, adds a short session and its own rules. Two tools learned well beat four learned badly.
- Low confidence with computers. Some capable staff are slow typists or rarely use office software. Pair them with a colleague and double their practice time rather than rushing them.
- Work that needs checking against source documents. Estimating, bookkeeping and anything with figures needs extra practice on verification, because AI makes confident arithmetic and quantity errors.
Three ways to deliver it
| Approach | Cash | Staff hours (10-person team) | Suits | Risk |
|---|---|---|---|---|
| All in-house with free courses | $0 to $50 | About 60 to 80 | Teams with a capable champion who has time | Uneven uptake if the champion is stretched |
| In-house plus one external session | Free courses plus one trainer's fee (quote) | About 60 to 80 | Teams that want an expert kick-off on their own tools | The session is good, then nobody follows up |
| External programme | Trainer's fees for several sessions (quote) | Roughly the same; less for the champion | Larger teams, or no obvious champion | Generic content if the trainer doesn't learn your tasks first |
Notice that paying more barely changes the staff hours. The practice time is the same whichever way you deliver the introduction, because practising has to happen on real work. What paying more buys is a better introduction and less load on your champion.
Ways to cut the hours without cutting what matters
- Train on three real tasks per role, not on "AI". Choose tasks each person does weekly, so practice happens during work rather than on top of it.
- Swap long sessions for short clinics. Twenty minutes a week where people bring a real problem beats a second half-day. Running a 60-minute AI workshop gives a format for the kick-off itself.
- Give people prompts that already work. A shared prompt library means nobody starts from a blank box.
- Record the introductory session. Your next new starter gets it for free, which is most of what onboarding new hires onto your AI tools needs.
- Pair people. An estimator who has cracked the quoting prompt can teach the other one in 30 minutes, which is quicker than any course.
To see why clinics are such good value, here's a typical one. An administrator at an illustrative letting agency brings a prompt she's been using for property listings, and a sample of what it produces:
HER PROMPT
Write a listing for this flat: 2 bed, 1 bath, second floor, new
kitchen, parking space, available 1 November, $1,650 a month.
ILLUSTRATIVE OUTPUT
"A stunning two-bedroom apartment in a highly sought-after location,
close to excellent schools and transport links, with a brand-new
designer kitchen and allocated parking..."The flaws: "highly sought-after", "excellent schools" and "designer" came from nowhere, and a tenant who relies on a claim about the area has a fair complaint. Ten minutes in the clinic produces two added lines: "Use only the features listed. Make no claims about the area, schools or transport. Plain, factual tone; no adjectives like stunning." The next output reads like the flat. That fix now sits in the shared prompt library, so the other two administrators never make the mistake, and the clinic cost 20 minutes of three people's time.
How to tell the training paid off
Check at 30, 60 and 90 days, and look for four signals. First, regular use: most business plans show active users in the admin console, so check who has actually used the tool in the past fortnight. Second, each person can show you two tasks they now do with AI, with the prompt they use. Third, time per task has fallen on the tasks you trained, measured the same way you measured it before. Fourth, error rates on AI-assisted work are flat or falling, because training that increases speed and mistakes together has failed.
For the caterer's sales coordinator, the day-60 check fits on a few lines:
DAY-60 CHECK: sales coordinator
Active days in the past 14: 9 of 10 working days
Tasks shown with prompt: draft quotes; menu descriptions
Draft quote, first reply to sent: was about 45 min, now about 20 min
Errors caught at manager check: 2 in 6 weeks (a wrong canapé price
in each; price list re-pasted weekly)
Task 3 (tasting feedback): not used; chef prefers notes as they
are. Dropped from her plan.Dropping a task nobody wants is a good result, not a failure. It frees the practice time for the two that are working.
If use drops off after week four, the cause is rarely the training content. It's usually that people weren't given time to practise, or the tasks chosen weren't ones they do often. If you sell to customers in the EU, keep a simple record of who was trained on what and when; it also supports the AI literacy measures described in what your staff need to know about AI literacy.
Training questions owners ask next
Do we have a legal duty to train staff on AI?
If you sell to customers in the EU, the EU AI Act's AI literacy article has applied since 2 February 2025. Its wording was softened by the Digital Omnibus in force from 27 July 2026: businesses must take measures to support their staff's AI literacy, without having to guarantee a particular level. Short, recorded training on your tools and rules is a sensible way to show those measures. Ask an adviser about your own obligations.
How often does training need repeating?
Full sessions rarely need repeating, but short refreshers do. Plan a 20 to 30 minute update whenever a tool changes noticeably, you add a new approved use, or your rules change, plus a brief yearly check that everyone still knows the data-handling rules. The weekly or fortnightly clinic in the first two months usually tapers naturally once questions dry up.
Are vendor certificates worth paying for?
For most small-business staff, no. A certificate shows someone finished a course, not that they use AI well on your work. The free vendor courses teach the same fundamentals. Paid certification makes more sense for the one person who will champion AI internally or advise others, and even then practice on your real tasks matters more than the badge.
Further reads
- How to Get Your Staff to Actually Use AI Tools — What to do when trained staff still don't use the tools.
- How to Learn AI as a Business Owner: A 30-Day Self-Study Plan — A 30-day plan for the owner's own learning.
- How to Survey Your Staff Before an AI Rollout (With Questions) — Find out what people already know before you plan hours.
- How Much Should a Small Business Budget for AI? — Fit training time into the wider first-year AI budget.
- How to Build a Staff Training Matrix With AI — Track who has been trained on which tool.
- How to Build a Shared Prompt Library for Your Team — The resource that cuts practice time for every new user.
- How Much Does AI Cost a Small Business in 2026? — Itemised 2026 AI budgets for a sole trader, a six-person office and a 20-person firm, with the charges that grow after month three.
- What AI Costs a Small Retail Shop in Its First Year — An itemised first-year AI budget for a small shop, with three worked bills, the month each cost lands and the costs that never appear on an invoice.
- How to Build a Kitchen Training Manual With AI — Capture how your kitchen really works, let AI shape it into station cards and quizzes, and keep every temperature and allergen under the head chef's control.
- What Does AI Cost a Small Insurance Brokerage in Year One? — Year-one AI costs for four, eight and twelve-person brokerages, quarter by quarter, including the staff time and compliance work most budgets leave out.
- Hire an AI Specialist or Upskill Your Agency Team? — When an agency should train the team it has, when a specialist hire pays for itself, and the hybrid most agencies under twenty people end up with.
- Hidden Costs of AI Implementation Most Small Businesses Miss — The AI costs that never appear on a pricing page, a subscription box company's real first-year bill against the headline, and a cost sheet to copy.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Claude Academy course catalogue (academy.claude.com); OpenAI Academy courses help page; Microsoft Learn Copilot training paths; Coursera, Google AI Essentials course page; EU AI Act Article 4 and the Digital Omnibus on AI.