Sam Wong.黃力桐Let’s talk training

TRAINING DECISIONS

Your in-house AI trainer's first 90 days: a one-page plan for HR

Published:

You have a few names. Maybe they volunteered after a workshop. Maybe their managers put them forward. Either way, someone in HR now owns the question: what do these people actually do for the next three months?

This is the plan I would hand you. It fits on one page, and there is a checklist at the end you can print.

Why 90 days

Ninety days is long enough to see behaviour change and short enough that nobody forgets why you started.

It also matches what I see after training. People are keen for a week or two. By the second month, unless something is in the calendar, work quietly goes back to how it was. That is why the programme for the College of Surgeons of Hong Kong had a third part: an online Q&A clinic about six weeks after the two evenings of training. I opened it like this (translated from Cantonese):

"Learning AI, or applying AI, is not a one-off thing. It's not that after one workshop everyone becomes an AI master, or uses AI really well. It's a long-term journey." — Sam Wong, CSHK Q&A Clinic, 5 October 2026 (01:00–01:21)

Research points the same way. In a 2026 study of a US state extension service, more than half of the staff surveyed (56.4%) named no one inside the organisation they could go to for AI advice. Your in-house AI trainers are meant to be that person. The 90 days are how they become one.

Before day 1: three decisions HR makes

The rules. Which AI tools are approved, and what data may go into them. If staff aren't sure, careful people do nothing, and that is the rational choice. Write it on one page with your IT: red, amber, green. I wrote more about staff AI data rules in Hong Kong.

The time. Agree with each trainer's manager, in writing, how many hours a week the role gets. Without that, the role becomes extra work, and extra work is the first thing dropped in a busy month. I wrote about what happens next in why your AI champions give up within a few months.

The people. A small group, with someone from each department. Pick people colleagues already go to for help, not only the most enthusiastic. Ask each one to bring one task from their own work that AI could help with, and time it now. That is your baseline.

Days 1 to 30: build on their own work

The first month is about their own use, not teaching yet.

Each trainer takes one real task from their department, a weekly report, a supplier email, a policy summary, and puts AI into it inside the approved tools. Then they time it again. One task, one before-and-after number.

That number is the first thing they will show colleagues, and it is more convincing than any demo I could run.

At Garden, each department formed its own cohort and practised on exercises built around its own work, with fictional data only. That is the pattern: the real shape of the work, with safe data.

Days 31 to 60: learn to teach it

Now each trainer designs one short session for their own colleagues, built on the task they just improved. It needs three things:

  1. The exercise itself, on a task everyone in the room recognises.
  2. A run-of-show with timings. The run-of-show matters more than the slides.
  3. A plan for when the AI tool fails in front of the room. AI tools break live. What you do in that moment decides whether you keep the room.

They rehearse it with me, then run it for real. I watch the first one and debrief afterwards.

This is why the first session should be on work everyone already knows. The colleagues in the room are the experts on the task. The trainer only has to show the AI part.

Open the question channel in this month too. At HKCT, the WhatsApp AI question group has more than 200 members. Yours will start smaller. What matters is that someone answers.

Days 61 to 90: run it without me

In the third month, each trainer runs at least one more session on their own, and the channel keeps going. A regular clinic, where colleagues bring whatever they got stuck on, gives everyone a date to work towards.

Keep a simple log: sessions run, colleagues who came, questions asked in the channel.

Day 90: count behaviour, not satisfaction

A satisfaction score tells you people liked a session. It does not tell you whether anyone opened an AI tool at their desk the next week.

So at day 90, count five things:

  1. Sessions your trainers ran.
  2. Colleagues they reached.
  3. Tasks now done with AI, with the before-and-after time.
  4. Questions asked in the channel.
  5. Whether another department asks to join.

If those numbers moved, keep the weekly hours in place and add the next group. If they didn't, you know in month three, not month nine, and you know which of the five numbers broke.

The one-page checklist (print this)

Before day 1

Days 1 to 30

Days 31 to 60

Days 61 to 90

Day 90 review

Want a 90-day plan built around your own tools and teams? I train a small group of your staff as in-house AI trainers and support them through the first 90 days. Tell me how many staff you have and which AI tools you're allowed to use. WhatsApp me · How the in-house AI trainer programme works

FAQ

What does an in-house AI trainer do in the first 90 days?

They put AI into one task of their own work, then design and run a short session for colleagues, then keep a question channel and a regular clinic going. HR checks five numbers at day 90.

How many in-house AI trainers does a company need?

Start with a small group, with someone from each department, so every department has a named person colleagues can ask.

What should HR measure after AI training?

Behaviour, not satisfaction: sessions run by your trainers, colleagues reached, tasks now done with AI, questions asked, and whether other departments ask to join.

Sources

Sam Wong (黃力桐) is a co-founder and Director of Academy at Adaptig. He designs AI training for teams and trains the trainers who deliver it.