AI Training for Staff: 8 Questions to Ask Before You Book

Why most AI training changes nothing by Monday

Most companies that have run an AI training can tell you how it felt. Fewer can tell you what changed. Four patterns explain most of the gap.

The inspiration day. A strong speaker, impressive demos, a room full of people nodding. On Monday they open their inbox and work exactly as they did last week, because nobody practised on their own work.

Generic examples. Marketing slogans for a finance team. Holiday itineraries for a customer service desk. People cannot see their own job in the exercise, so they file the session under "interesting".

The wrong tool. The trainer teaches a favourite assistant. The company licenses a different one. On Monday half the room cannot find the button they learned, and the other half quietly signs up for a personal account. That last part is a data problem, not only a training problem.

Nothing afterwards. No follow-up, no owner, nobody checking whether anyone still uses what they learned. The new habit fades within weeks.

None of this means the trainer does not know AI. These are design problems, and you can spot every one of them before you sign.

What does good AI training for staff involve?

A training that changes how people work usually has these parts:

• A survey before the session. Roles, current AI use and the written tasks that take the most time. The exercises are built from the answers.
• Your own document types. Real emails, reports, offers, meeting notes and spreadsheets, with names removed.
• The tool you already license. Microsoft Copilot, ChatGPT, Claude or Gemini: whichever your company pays for. The others appear only for comparison.
• A data rule people remember. What may go into which tool, and which account settings to change. We teach a traffic light rule; our one-page AI policy template sets it out in full.
• One shared prompting method. So the whole team talks about prompts the same way. Ours is ROCSRS.
• Something that stays. A one-page cheat sheet, the prompts built in the room, a recording if you want one.
• A follow-up. A check two to four weeks later, and a named owner for each use case worth keeping.
• A written summary for management. What was covered, how participants rated it, and which workflows are worth taking on next.

In our workshops more than half of the time is hands-on, with people typing in their own accounts on their own tasks. We keep it to at most three tools live in one session. Beyond that, people remember the menus and forget the method.

8 questions to ask before you book AI training

Send these to every trainer on your shortlist. The answers will tell you more than the brochure.

#	Question	A good answer sounds like	Red flag
1	Will we work on our own documents?	"Send us three typical document types. We build the exercises from them, with names removed."	"We have a standard set of examples that works for every team."
2	Which tool will you teach, and is it the one we license?	"The one you already pay for. We show the others only for comparison."	One favourite tool for everyone, or advice to switch before they have seen your setup.
3	What happens before the session?	"A short survey: roles, current use, the most time-consuming written tasks."	Nothing beyond a date and a headcount.
4	What data rules will people leave with?	A simple rule for what goes into which tool, plus the account settings to change.	"Just be careful with sensitive data."
5	How much of the time is hands-on?	More than half, in participants' own accounts.	Three hours of slides and a Q&A at the end.
6	What stays with us afterwards?	A one-page method, the prompts built in the session, a cheat sheet, a recording on request.	The slides as a PDF.
7	What happens after the training?	A follow-up, named owners for the first use cases, a written summary for management.	"Call us when you are ready for the advanced course."
8	Who is the trainer, and what have they actually built?	Assistants and workflows in daily use at clients, including what broke and how it was fixed.	Certificates, keynotes and follower counts, but no working examples.

Three of these deserve a closer look.

Question 2: the tool decides what happens on Monday

If your company runs on Microsoft 365, a training built on ChatGPT teaches people a tool they may not be allowed to use for work. The method transfers; the buttons, limits and data settings do not. A good trainer asks what you license first and adapts the session to it. If you have not chosen a tool yet, ask for a short comparison on your own tasks rather than a recommendation from slides. Our tool-specific pages cover Copilot, ChatGPT and Claude.

Question 4: data rules are the part people actually use

One question comes up early in our sessions, and it has nothing to do with prompts: "Can I put this in?" People need an answer they can apply in two seconds. Ours has three colours. Green information can go into any account: product information, drafts without names, translations, spreadsheet formulas. Yellow goes only into the company account, and a person reads the result before it is sent: client emails, orders, supplier letters, complaints. Red goes nowhere: passwords, bank access, employee personal data and salaries. One line runs through all three: nothing beyond green ever goes into a personal account. If a trainer cannot give your team something this concrete, people will invent their own rules.

Question 7: the first weeks after training decide

A new habit either forms or fades in the weeks right after the session. A follow-up two to four weeks later, even a short one, shows who kept using what and where people got stuck. Naming an owner for each use case worth keeping is what turns a good day into a working practice. We wrote more about that in why most AI adoption efforts fail.

What drives the cost of AI training?

Prices vary a lot, and comparing per-person rates hides most of what matters. These factors move the cost:

• Group size and number of groups. One workshop suits a group of 10-50. Larger organisations are split into several groups, often by function.
• Preparation and customisation. Analysing the survey and building exercises from your documents takes time. It is also what makes the session relevant.
• Language. English, Estonian, or materials in both.
• Online or on site. Online is simpler to organise and to record; on site adds travel.
• Follow-up. A single session costs less than a workshop with follow-up sessions or group work afterwards.

One number is worth working out yourself: the hours your people spend in the room. Fifty people in a four-hour session is 200 working hours. That is the real investment, which is why relevance matters more than the rate.

Free online courses are a good way for one person to learn the basics at their own pace. A company training solves a different problem: the whole team learns the same method and the same data rules, in the tool you actually use, on your own tasks. If the goal is a shared way of working, an individual course will not replace it.

Stairway to AI does not publish prices. After a discovery call you get a fixed price in a written proposal within 48 hours.

One workshop or a programme?

A single half-day workshop is enough when:

• you want a shared starting point: one method, one data rule, one vocabulary;
• people already use AI on their own and need to align;
• you want to see how a trainer works before committing to more.

A programme makes more sense when:

• you want AI built into specific workflows, not just better individual use;
• each team should leave with a working assistant for a recurring task, with a named owner;
• management wants a tool and licence decision based on how people actually used the tools.

Our AI Onboarding Partner programme runs over 2-3 months in three stages: a foundations workshop for everyone, deep-dive sessions in which each group builds a working assistant, then implementation and mentoring with a written summary for leadership. The AI adoption plan for the first 90 days walks through the same arc for teams that want to run it themselves.

Can beginners and advanced users train together?

Yes, and mixed rooms often work better than people expect. The survey shows the spread. The exercises then have levels: beginners simply ask, the next level writes a structured prompt, and the most experienced let the AI write a reusable prompt for a recurring task. Advanced users get a harder problem instead of a slow morning, and they become the people others ask afterwards.

What does not work is mixing functions that share nothing. Finance and sales can share the foundations session, but their deep-dives should be separate, because their documents are.

What to do this week

Next 15 minutes. Write down the three document types that take your team the most time to produce. Check which AI tool your company already licenses, and who has access to it.

Next 7 days. Send the eight questions above to the trainers on your shortlist. Ask each one how they would build an exercise from one of your three document types. Compare the answers side by side.

Next 30 days. Book the session, and book the follow-up at the same time. Decide now who will own the first use cases afterwards.

Sources

Regulation (EU) 2026/1744 amending the EU AI Act (Digital Omnibus on AI), EUR-Lex. Product and legal facts checked on 29.10.2026.