In Practice
Apply AI to real work, and check every output before you use it.
Everyone learns the same core skills, then applies them to their own work.
A finance manager, a facilitator, and an operations lead do not need three different AI trainings. They need the same footing — how to instruct it, how to check it, how to keep data safe — and then a thin layer of their own real work laid on top. About seventy percent is the shared core; the remaining thirty percent applies AI to each person's actual job.
This is what keeps a rollout coherent. When the foundation is common, people can help each other, share what works, and grow along one ladder instead of scattering into private habits no one can see or reuse.
It is fast and widely read, and it drafts in seconds what would take you an afternoon. It also states wrong things with full confidence. So treat it as you would a capable new hire: give it clear instructions, and check its work every time. It speeds up the work, but it does not make the final decision.
Always check. AI can be confidently wrong — inventing a figure, a citation, or a policy that sounds right. Read every output before you rely on it, and never let it be the last set of eyes on anything that matters.
Give clear instructions. Vague in, vague out. The more you tell it about the situation, the task, and the shape you want back, the better the result. Instruction quality is the single biggest lever you control.
Keep it confidential. Company data belongs in the company's sanctioned AI workspace — never in a personal account or a random consumer app. Where the data lives decides which tool you are allowed to use.
Almost every good prompt has three parts. Context — who you are, what the situation is, what the AI needs to know. Task — the one thing you want it to do. Format — the shape you want back. Miss one and the output drifts; supply all three and it lands close on the first try.
Worked example: “I manage the front desk at a serviced-apartment building and a guest has emailed to complain that their aircon was noisy for two nights (context). Draft a warm, apologetic reply that offers a partial credit and a follow-up call (task). Keep it under 120 words, friendly but professional, and give me two subject-line options (format).” — then read it, correct the details only you know, and send it yourself.
Notice the last move: you still read and correct before it goes out. The formula gets you a strong draft, not a finished decision.
AI is only as safe as the boundary you keep around it. The rule is simple: ordinary working material inside a sanctioned company workspace is fine; anything that identifies a person financially or personally does not go in at all.
When in doubt, leave it out — or summarise it so the sensitive specifics never reach the tool. And always use the company AI for company data, never a personal account.
The safe workflow is always the same three steps: AI drafts, you read and correct, then you send. The AI never presses send, and its draft is never the final word. This one habit prevents almost every embarrassing mistake.
Two more habits that compound quickly. Summarise instead of forwarding. Rather than forwarding a screenshot or a long thread, ask AI for a tight summary of what it says and what it asks for — turn images and clutter into a typed answer. And translate on demand — a Thai email you need in English, or an English document a colleague needs in Thai, is now a few seconds of work, not a bottleneck.
Apply AI to real work, and check every output before you use it.
Gemini Gems, Claude Projects, or a custom GPT — the same idea by different names. You save the setup once so you never re-type it: a fixed role, the context it always needs, the task it usually does, the format you want back, its limits, and a self-check rule. A good helper always ends by flagging what it was unsure about — and you still check. It saves you from re-briefing the AI every time you do the same job.
Most of the value AI creates goes unrecorded: a draft that saved an hour, a summary that avoided a meeting, a translation that unblocked a decision. None of it is recorded anywhere by default. The weekly report records it.
It starts almost trivially small — three questions, answered once a week:
1. What did AI help me with this week?
2. Did it actually work?
3. What will I try next?
That is all it is, at first. Over the weeks it grows — which prompts worked, which helper you built, where AI saved real time — until it is no longer a training exercise but a running record tied to your actual job. The weekly report is the core of the curriculum: it turns scattered experiments into visible, reusable, reportable progress.
A running record of what AI did each week, so progress is something you can see.
Progress with AI is not pass or fail. It runs from never having used it to building tools the whole team reuses. The ladder shows where each person is today and what the next step is, and lets a rollout track its own progress.
Not yet using AI in any visible way.
Understands what AI is and why it matters for the work.
Has applied it to actual work, not just a demo.
Can show what was asked and what came back — and treats it as a draft.
Reads, corrects, and owns the result before it goes anywhere.
Saves the setup so a recurring job runs the same way every time.
Maintains a running record of what AI is doing across their real work.
Turns personal practice into something the whole team can reuse.
The foundation is taught as a progression — each step small enough to do this week, each one building on the last. It ends with AI doing real work on your actual job, with a written record of it.
The one non-negotiable across every step: it must touch your actual job. The goal is not a tidy summary about AI but a draft you actually sent, a helper you reuse, and a report tied to work you own. Work that never touches a real task does not count.
One shared foundation for the whole team, applied to each role.
Request a consultation