The HR Adoption Handbook

Before you begin

The job is not to teach AI. It is to make it safe and normal.

Most AI rollouts stall for the same reason: nobody decided the boring things. Which tool, who can use it, what they must never paste into it, and who to ask. Settle those, and the training becomes easy.

The one idea to repeat to everyone

AI gives you the speed. You keep the judgment. Check every fact before it leaves your hands.

This handbook is a starting point you adapt to your company. It is not legal advice, so have your counsel or compliance team review the policy before you publish it.

01 · Decide

Five decisions to make before anyone is trained

Write the answers down. One page is enough.

Which tools are approved?Name them. If your company already pays for Microsoft Copilot, Claude or Google Gemini, start there. People will use something anyway, so give them an approved option. Ask your IT team what protections your plan includes, and write down what they tell you.
Who gets access first?Start with one team that has repeatable work and a manager who is curious. A small group that succeeds is the best advertisement for everyone else.
What must never go in?Settle this in writing (section 02). It is the single decision that prevents most problems.
Who answers questions?Name a person, not a committee. People stop using a tool when they do not know whom to ask.
How will you know it is working?Pick two or three signals now (section 06), so you are not guessing in three months.

02 · Protect

What data is safe to use

Give people a simple traffic light. They will use it, because it is easy to remember.

Level	Examples	The rule
Green	Public information, your own drafts, general questions, made-up examples	Fine in any approved tool.
Yellow	Internal documents, customer names, pricing, meeting notes	Only in the approved company tool, signed in with a work account. Remove what you do not need.
Red	Passwords, personal health or financial details, employee records, anything under a contract or law that restricts it	Never. Ask your compliance contact first.

Scrub it first. Swap real names, account numbers and amounts for placeholders when you only need the pattern. It takes a moment and lowers the risk.

Your own industry may add rules of its own, for example for health, legal or financial work. Check with the person responsible for compliance before you lock the table.

03 · Write it down

The one-page usage policy

Copy this, fill in the highlighted parts, and have your counsel review it. Short policies get read. Long ones get ignored.

How we use AI at Company name

Approved tools. List the tools, signed in with your work account. Please do not use other AI tools for company work.

What you may do. Draft, summarize, brainstorm, analyze and organize your own work.

What stays out. Red-level data, as listed in our data guide. If you are not sure, ask Name or role first.

You are the author. You are responsible for anything you send or act on, whether or not AI helped. Check facts, numbers, names and quotes against a real source before they leave your hands.

Be open about it. Tell your manager how you use it. Where a client or customer would reasonably want to know, say so.

Questions and mistakes. Ask Name or role. If something goes wrong, tell us early. We would rather fix it together than find it later.

Reviewed: Date. We review this policy every 6 months, because the tools change.

04 · Roll out

Your first 90 days

Small, steady and visible beats a big launch that fades.

Days 1 to 30: prepare and pilot

Make the five decisions and publish the one-page policy.
Choose one pilot team and one manager who will go first.
Give the pilot team the training that matches their tool (section 05) and ask each person to pick one repeat task to try.
Hold a 20-minute kickoff. Show the policy, show one real task done with AI, answer questions.

Days 31 to 60: learn what works

Meet the pilot team every two weeks for 20 minutes. Ask: what saved time, what went wrong, what surprised you?
Collect the prompts that worked into one shared document.
Fix the gaps you find: unclear rules, missing access, tasks the tool is wrong for.

Days 61 to 90: widen it

Share two or three real before-and-after stories from the pilot, in the team’s own words.
Add the next team, with the pilot team members as their first helpers.
Add the training to new-hire onboarding, so every new person starts with the same rules and the same basics.

05 · Train

Matching people to training

Different people need different starting points. Match the training to the tool they will actually use and the work they actually do.

If your team…	Start with
Works mostly in Microsoft 365 (Outlook, Word, Excel, Teams)	The Microsoft Copilot Academy
Uses Claude for writing, analysis and building tools	The Claude Academy
Uses Google Workspace	The Gemini Academy
Uses Meta AI	The Meta AI Academy
Needs a general library that covers many departments	The Vinyl Academy: Rock & Roll
Does finance, data or IT work	The Trading Desk, The Data Desk, or The Geeky Nerds
Runs the systems, security or helpdesk	The Control Room

Each person works through their academy at their own pace, starting with Step 1. Every seat is one person with one login, so each person you train needs their own purchase. Email us for multi-seat arrangements.

Be honest about what the platform tracks. Each person’s progress is saved in their own browser, so the academies do not send completion reports to HR. Use a short check-in conversation instead (section 06).

06 · Measure

How to tell whether it is working

Measure what people tell you and what they make, not what a dashboard guesses.

Ask three questionsEvery month, ask each pilot member: What did you use it for? About how much time did it save? What did you have to fix?

Count repeat tasksHow many routine tasks now have a saved prompt or a standard way of doing them? That number only goes up when people trust the tool.

Watch for qualityAsk managers whether the work that comes back is as good, better or worse than before. Fewer rewrites is a real signal.

Listen for questionsGood questions mean people are trying. Silence usually means they stopped. Go ask them.

Keep your own numbers modest and honest. A rough estimate from the person doing the work is more useful than a precise figure nobody believes.

07 · Talk

The hard conversations

People are rarely against AI. They are worried about something specific. Answer that.

“Is this going to replace me?”Say what you know: you are asking people to use it to do their own work faster, and the judgment stays with them. Do not promise what you cannot control. Do offer to revisit the question openly as you learn more.

“What if it gets something wrong?”It will, sometimes. That is why the rule is that you check before it leaves your hands, and why reporting a mistake early should be treated as the right thing to do.

“Is my data safe?”Point to the traffic light and the approved tool. Be specific about what you have confirmed with IT and what you have not.

“I am not technical.”Good. If you can write an email, you can do this. Start with one small task, then a second.

A script for a manager’s first team conversation

Say something like this

We are going to try this on one task each, for a month. You choose the task. You decide what goes in, within the rules. You check the result before it goes anywhere. Tell me what works and what does not, and nobody gets in trouble for either.