How to Choose an AI Training Provider for Your Team

Start by writing down what each role should be able to do after the training, using your own tools and tasks. Then ask two or three providers to quote against that list, test each syllabus for out-of-date product names, compare quotes per useful learner-hour including staff time, and pilot one short session before booking everyone.

The biggest cost of most small-business AI training isn't the provider's fee; it's the hours your staff spend away from their work. Three hours for fourteen people is 42 hours of wages and cover. That's why a generic "introduction to AI" that only half applies to your team is expensive even when it's cheap, and why the choice depends more on fit than on price.

Follow me on Instagram@sagnikteaches

Outcomes by role, written before any provider is contacted

The example throughout is an illustrative bakery with 14 staff: six bakers on early production shifts, five shop staff across two shops, two in the office and the owner. The office uses Microsoft 365 with the free Copilot Chat; the shops answer Instagram and website messages from a tablet. The owner's first step was a table of what each group should be able to do, in their own work, a month after training:

Connect on LinkedInSagnik Bhattacharya

The table immediately shows the bakers don't need what the office needs. A single all-staff course would either bore six people or under-serve two. It also puts the allergen rule and the data rules in writing before anyone designs a session, which is what a good provider will ask for anyway. If you haven't written those rules yet, rolling out an AI policy staff follow covers them.

Subscribe on YouTube@codingliquids

Five kinds of provider, and when each fits

AI training providers come in five broad forms, and the right one depends on who is learning what.

The free options are better than many owners expect. OpenAI Academy offers free, self-paced courses to anyone with a ChatGPT account, each with an assessment. Anthropic's Claude Academy has free AI Fluency courses built around delegation, description, discernment and diligence. Microsoft Learn has free Copilot learning paths, and Microsoft 365 users can reach Copilot Academy through Viva Learning. None of them knows your allergen rules or your suppliers, which is exactly the gap a paid provider should fill.

For the bakery, that suggested a mix: a free vendor course as homework for the office and owner, a short live session per group on the bakery's own tasks, and nothing longer than 45 minutes for the bakers. Doing the whole thing in-house is also an option; training staff to use AI yourself covers that route.

The out-of-date syllabus test

AI products change monthly, and course material lags. Ask each provider for the syllabus and the date it was last revised, then look for names that have changed or disappeared. Several did in 2026 alone:

• "Build a custom GPT for your team." OpenAI is retiring custom GPTs; they stop running on 11 December 2026. Current courses teach ChatGPT Projects for shared context, or Claude Projects and Gemini Gems (which become skills from November 2026).
• "Automate tasks with ChatGPT agent." ChatGPT agent was retired on 9 July 2026 and replaced by ChatGPT Work.
• "Use the =COPILOT() function in Excel." Microsoft retired that worksheet function on 14 September 2026; the Copilot side pane replaces it.
• "Research with NotebookLM." Google now calls it Gemini Notebook.
• "Zoom AI Companion." Zoom retired that brand in June 2026; the features are now named for what they do.
• "Make marketing videos with Sora." OpenAI discontinued Sora's app and API in 2026.

An outline with one of these isn't necessarily a bad course, but it tells you the provider isn't revising often. Ask what they'd teach instead and when the material was updated. A provider who replies with a revised outline the same week is showing you how they'll keep your staff current. One who says "the principles are the same" is telling you they won't.

A training exercise worth insisting on

The best sessions use your real situations, including ones where AI gets it wrong. Ask each provider how they'd teach this, and judge the answer. The shop staff's exercise at the bakery started with this prompt:

A customer has messaged us on Instagram: "Are your brownies gluten
free? My daughter is coeliac." Draft a friendly reply from the shop.

The draft a chat assistant produced (illustrative) read: "Hi! Yes, our brownies are gluten-free, so they're perfect for your daughter. Pop in any time!" The tool had no idea what's in the bakery's brownies. It produced the friendliest plausible answer.

What the trainer taught from it: the AI invented a fact; even a gluten-free recipe made in a kitchen that handles wheat flour can't be promised as safe; and allergen questions go to the written allergen sheet and a person, every time. The corrected reply read: "Thanks for asking. Our brownies are made in the same kitchen as our wheat-flour products, so we can't guarantee any of our bakes are free from gluten. Please call the shop and we'll talk you through what's safest." Staff remember an exercise like that. They don't remember a slide saying "check AI output". For the shop-floor angle in more depth, see training front-of-house staff to work alongside AI.

The quote request that makes providers comparable

Send the same brief to each provider, with the outcomes table attached. Something like:

Subject: AI training quote for a 14-person bakery

We'd like AI training for four groups (table attached): office (2),
shop staff (5), bakers (6) and owner (1). We use Microsoft 365 with
Copilot Chat; shop staff answer messages from a tablet.

Please tell us:
1. Format, length and timing for each group (bakers finish at 11am)
2. Which of our own tasks you'd use as exercises
3. When your material was last revised, and your current outline
4. How you'd cover our data rules and allergen rule
5. What follow-up you include after the sessions
6. How you'd measure whether it worked
7. Whether you sell or earn commission on any AI tool
8. Your total price, and what's excluded

Question 7 matters because some trainers are also resellers or partners of a particular tool. That's fine, but a trainer who earns from ChatGPT seats has a reason to steer you towards them even though your office runs on Microsoft 365. Question 1 matters for a practical reason: training scheduled across the bakers' shift pattern is training they'll miss.

Comparing three quotes per useful learner-hour

The bakery received three proposals. Prices aren't shown, because course prices vary widely and you should compare your own; what's shown is everything else you need to turn a price into a fair comparison.

Then put a price on it with this line: (provider fee + wages for training hours + cost of cover) divided by useful learner-hours. Wages are the part owners forget. At an illustrative $18 an hour, proposal A's 49 hours cost $882 in wages before the fee, and many of those hours go on examples nobody uses. Proposal B takes 19.5 hours, about $351, for almost the same number of useful hours. Unless B's fee is dramatically higher, it wins on the sums as well as the fit. For the wider picture of what staff training costs in time and money, see how much time and money AI training takes.

What a 45-minute session for the bakers covered

"Tailored" is easy to promise, so ask to see a session outline before you book. Provider B's plan for the bakers, who start at 4am and have the least use for AI at work, fitted into the quiet half hour after the morning bake plus a quarter of an hour of questions:

1. Five minutes: where AI is used in this bakery. The office drafting supplier emails, the shops drafting message replies, and the owner's production planning. Nothing about the history of AI.
2. Ten minutes: the three rules. No customer or staff details in AI tools; only the business accounts; allergen and recipe questions never answered by AI.
3. Fifteen minutes: one exercise. A chat assistant scales the sourdough recipe from 20 loaves to 55. The bakers check its figures against their own maths and find the salt quantity rounded wrongly. It's the fastest way to show why checking matters, using numbers they know by heart.
4. Fifteen minutes: questions. The trainer notes anything the owner should add to the written rules.

A provider who can produce an outline like this from your outcomes table has understood the job. One who sends the same agenda for every group hasn't.

Proposal claims to treat with caution

• Promised productivity percentages. "Staff will be 40% more productive" has no basis for your business. Ask how they'd measure it in yours instead.
• Certificates as the outcome. A completion badge shows someone finished a course, not that they check their work. Measure a task, not attendance.
• Everything on the trainer's own accounts. If exercises can't run on your tools and settings, staff learn habits you'll have to undo.
• "Unlimited access" to a video library. Useful for a keen few, rarely opened by the rest. Price it as an extra, not the core.
• No mention of your rules. A proposal that never asks what staff mustn't do with AI will skip the part that protects the business.
• A long commitment. Tools change quickly; a year-long programme bought upfront may teach last year's products. Prefer a pilot session, then a decision.

A mistake that comes from the wrong training

Here's a realistic way generic training backfires (illustrative). A business books an energetic webinar that demonstrates everything on the presenter's personal ChatGPT account. Afterwards, three staff sign up for free personal accounts and start pasting customer orders and supplier emails into them, because that's what they saw. Nobody broke a rule on purpose; the training simply never mentioned that the business had chosen Copilot Chat under its own Microsoft 365 accounts, or why.

Business plans such as Microsoft 365 Copilot, ChatGPT Business and Claude Team don't train on business content by default; personal accounts rely on each user switching off the model-training setting. Training that isn't built around your tools and rules can quietly move your data into places you didn't choose. That's the practical reason for insisting on your tools in every exercise, and for giving staff a shared set of approved prompts afterwards, as in building a shared prompt library.

Piloting one session, then measuring the rest

Before booking all four groups, the bakery ran provider B's office session first. It gave the two office staff the same practical check before and three weeks after:

The figures are illustrative, but the method is the point: a short practical check before and after, on your own tasks, tells you more than any feedback form. The office session passed, so the shop and bakery sessions went ahead. If it hadn't, the bakery would have lost one session's fee rather than four.

When you won't need a provider at all

A six-person car repair garage with one office computer and a service adviser who drafts customer emails probably doesn't need to buy training. The adviser can take a free vendor course matched to the garage's tools, the owner can run one 40-minute session on the garage's own rules (no customer card details, no promises about repair prices from AI drafts), and a one-page prompt sheet covers the rest. Buying training makes sense when several roles need different things, when the stakes of a mistake are high, or when nobody in the business has the time or confidence to lead it.

Records worth keeping after the training

Keep a simple record of who attended what, when, and what it covered, plus the materials. It helps when new starters join, when you retrain after a tool changes, and when anyone asks how staff were prepared. If you sell to customers in the EU, the EU AI Act's Article 4 has applied since 2 February 2025; since the Digital Omnibus on AI came into force on 27 July 2026, it asks businesses using AI systems to take measures to support their staff's AI literacy, without requiring a specific level. A dated attendance list and the session outlines are a sensible way to show the measures you took. What AI literacy requires of your staff covers the obligation in more detail; for anything beyond that, ask a qualified adviser.

Further reads

• How to Build a Staff Training Matrix With AI — Track who has learned what, role by role.
• How to Get Staff Buy-In When You Introduce AI — Get the team on side before anyone sits in a session.
• How to Onboard New Hires Onto Your AI Tools and Rules — Keep new starters trained once the course is over.
• Does a Five-Person Business Really Need an AI Policy? — Write the rules the training should teach.
• How to Train Staff to Spot AI-Written Phishing Emails — A security topic worth adding to any AI course.
• How to Stop AI Tools Training on Your Business Data — The settings staff should learn in the first session.
• 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.
• Course, 1:1 Call, Audit, or Project: Which AI Help Do You Need? — What a course, a 1:1 call, an audit and a project each leave you with, five questions to find your starting point, and one deli's route through three.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: OpenAI Academy pages and OpenAI help pages (Academy courses; custom GPT retirement; ChatGPT Work); Claude Academy course pages; Microsoft Learn and Microsoft Adoption training pages; Microsoft Support (Excel COPILOT function); EU AI Act Article 4 as amended by the Digital Omnibus on AI. Checked September 2026.