For: team leads at Helios (Finance, HR, Ops-planning) · Time cost: one 5-minute huddle per adoption stage + habit coaching · Rule of thumb: your team copies your behavior, not your memo.
Pick ONE recurring task per team. Not ten.
Teams that adopt ten AI use cases at once adopt zero. The playbook:
Your job as lead: use the habit yourself, visibly, including the failures. “The draft missed the VAT lines — here’s how I caught it” is worth ten policy emails.
“This month we’re learning to use AI tools on real work — starting with [team task]. Two things up front: the tools are good at drafts and bad at facts, so checking stays human work. And nothing with customer data, unreleased financials, or HR case details goes into an external tool — that’s a hard line, not a guideline. Demo on Thursday, bring your most annoying recurring task.”
Your move: collect the “most annoying task” answers — that’s your habit shortlist.
“Two people will show something real they made with a prompt pattern this week — one win, one mess. The mess is the important one: where did the draft go wrong, and how did you catch it?”
Your move: go first with your own mess. If nobody volunteers after you, the safety test failed — rerun the no-blame framing from Kit 3 before continuing.
“Our team habit is now [task]. Quick round: who used it, what did it save, where did it wobble? If you didn’t use it — what blocked you? Blockers are the agenda, not attendance.”
Your move: write blockers down and kill one per week, visibly. Blocker removal is adoption work.
“Three norms we now run by default: client-facing text drafted with AI gets a human name on it, always. Red-data never enters external tools — when unsure, it’s red. And near-misses get reported the same day, no consequences, because a reported near-miss is a free lesson.”
Your move: name your team’s champion and the escalation path. End with: “Questions you’re afraid to ask are exactly what the champion is for.”
| Never | Why |
|---|---|
| Customer/client PII (names + any identifying detail) | AVG/GDPR: no lawful basis for external processing; one paste is one incident |
| Unreleased financials, pricing, bids | Confidentiality + market-sensitivity; also a fast way to train someone else’s model on our numbers |
| HR case details (sickness, performance, disputes) | Special-category data; also destroys trust if it ever surfaces |
| Credentials, keys, internal system details | Security, full stop |
| Anything you’d panic about on the front page | The newspaper test is a legitimate control |
| Generally safe | With the habit |
|---|---|
| Anonymized/aggregated work content (patterns, structures, drafts) | Strip identifiers first — “a client” not “Vermeulen Transport” |
| Public information, generic drafting, formatting, summarizing non-sensitive text | Still: human checks facts before send |
| Your own task descriptions for prompt-building | The trainer’s sandbox is built for exactly this |
When unsure: treat it as red and ask the champion. Uncertainty resolves upward, never downward.
Fictional disclosure: Helios Facilities Group is a fictional client for this portfolio. Pack drafted with AI assistance; final judgment and copy are human-owned.
Prefer the source? Download the raw markdown.