# Before buying AI training, ask to see an exercise

> Comparing corporate AI training in Hong Kong? Ask the provider for one complete exercise, then check the assessment criteria, preparation, support and how outcomes are measured.

- URL: https://hyperfocusam.com/blog/five-questions-corporate-ai-training-hong-kong/
- Language: en
- Author: Sam Wong (黃力桐), https://hyperfocusam.com/
- 中文版: https://hyperfocusam.com/blog/five-questions-corporate-ai-training-hong-kong-tc/ (Markdown: https://hyperfocusam.com/blog/five-questions-corporate-ai-training-hong-kong-tc/index.md)
- Published: 2026-07-29; revised 2026-10-12

---

A course outline can list plenty of tools and sound tailored to your industry. But the outline alone tells you very little about what your colleagues will actually be able to do on their own afterwards.

If you are comparing corporate AI training proposals, I suggest asking for one complete classroom exercise: the material learners receive, the work they must produce, and the criteria the trainer uses to assess it. This gives you a concrete way to see if the “hands-on” training actually delivers.

## Define the work you want people to do

Introducing colleagues to AI is a reasonable goal. An executive briefing, an introductory workshop and a departmental practice session can all serve different purposes. The problem arises when the purchase is framed as an introduction, but you expect the result to be a change in how people work.

Try describing the need as something observable. For example: “When purchasing colleagues receive several quotations, they can prepare a comparison, identify missing information and draft specific follow-up questions.” This gives the course design a clear starting point. What information is needed? Which judgments remain with people? How much should learners practise the task?

A provider might not have a complete answer in the first meeting. However, they should be able to explain what they need from you to design the course properly.

## Look at the assessment criteria alongside the exercise

Take supplier comparisons. A learner gives several files to AI and receives a fluent analysis. The exercise could end there. Or it could require learners to identify the source of each number, spot unconfirmed payment terms, and distinguish the lowest unit price from a quotation that meets purchasing requirements.

The second approach makes the judgment being taught visible to the buyer. It also gives the trainer a basis for explaining what is usable and what still needs revision.

Ask to see these three items together, rather than just the most polished finished example.

| Material | What to look for |
|---|---|
| The input learners receive | Does it resemble the target task? Are there gaps or inconsistencies they must handle? |
| A sample output | Can its claims be traced to sources? Does it present assumptions as facts? |
| The trainer’s assessment criteria | What counts as completion? How will learners receive feedback when something is wrong? |

A provider who cannot share client information can offer a fictional case. It should be clearly labelled and preserve the judgments the task requires. A recognisable client name is not necessary for a good exercise.

## Check the timetable with 90/20/8

When a provider sends you their full exercise and the session timetable, you can check it against the rule I use to design my own workshops. This video shows how to apply my 90/20/8 rule and how to see if the session ends with something practical. It turns the vague question of whether a course is hands-on into a simple check any manager can do in five minutes.

Put the provider’s timetable next to you and check three numbers.

- 90 — Is there a real break (not just a pause) every 90 minutes?
- 20 — Does any single content block (lecture, explanation) run longer than 20 minutes before the format changes?
- 8 — Does the type of interaction change at least every 8 minutes: demo, discussion, exercise, sharing?
- Finally, does the session end with something tangible people take back to work, for example a checklist, a set of prompts, or a clear next step?

If a timetable fails these checks, it does not mean the course is useless; it means you should ask the provider why it is arranged that way and how people will stay engaged. If there is no timetable at all, ask for one.

## “We have Copilot” does not complete the preparation

Before learners practise, there is another check: can they perform the required steps on their own accounts and devices?

Ask the provider to list the features, files, accounts and steps involved. Your IT team can then confirm the available functionality and permitted data use. If someone cannot access a feature, will they receive an alternative exercise, work with a partner or watch a demonstration? These arrangements affect the objective you originally set.

The buyer does not need to become a technical expert. Access and permission questions need an owner before the course starts. Ask for a preparation checklist that also makes the organisation’s responsibilities clear.

## Specify what support will actually deliver

“Post-course support” might mean a handout. It might mean a trainer reviews work that participants have attempted. When comparing proposals, ask how long support lasts, which channel it uses, what questions it covers, who responds and whether feedback on work is included.

If the objective includes applying AI after the course, select a small task with approved data use and agree on a follow-up review. That review could cover the document the learner produced, unresolved problems, and help needed from a manager or IT. Arrange the timing around the work; not every team needs the same number of weeks.

These are deliverables you can agree on during procurement. Whether colleagues continue using the approach, and whether the quality of their work changes, requires separate observation afterwards.

## Ask what an outcome measure actually measures

Satisfaction, independent exercise completion, later usage and work outcomes answer different questions. A high satisfaction score tells you how respondents rated an experience. It does not, by itself, tell you how many people were still using the skills a month later.

When a provider gives you a result, ask what question was measured, when it was measured, how many people responded and which programme it concerned. If the claim is time saved, does that include checking and revision time?

A lack of long-term tracking does not mean a course has no value. The provider should explain what the evidence supports. You can then decide whether follow-up measurement belongs in this purchase.

## A message you can send to a provider

> We are evaluating corporate AI training to help our ___ team complete ___. Please provide a relevant exercise, including the input material, expected output, assessment criteria and feedback approach. Please also specify account and feature preparation, what our organisation needs to arrange, and the scope of post-course support. If you include outcome figures, state the programme scope, measurement method, timing and number of respondents. Fictional or de-identified cases are welcome; please label them clearly.

## 3 things you can do tomorrow

1. Write down one task you want the team to be able to do after the course, as an observable action (who receives what, and what they produce).
2. Copy the message above, send it to each provider, and add one line asking them to attach the session timetable.
3. When the materials come back, spend about 5 minutes per provider: check the exercise and assessment criteria, then check the timetable with 90/20/8; note anything missing and ask about it.

Once you have these materials, comparing quotes and arrangements is much easier because you can see what each proposal actually includes. If you ask me about training, tell me the work you want colleagues to do, and we can design the exercises and support together.