Your One-Page AI Policy (and How to Judge a Vendor)

Lesson 9 · taught by Pip · free preview

You've now toured the whole coastline — honesty, data, accuracy, guardrails, and the rights customers hold. This last lesson turns all of it into two practical things you'll actually use: a one-page policy for your business, and a short checklist for judging any AI tool someone tries to sell you.

Your one-page AI policy

A policy doesn't need a lawyer's letterhead to be useful. A single page that you and your team actually follow beats a fifty-page document nobody reads. Write down clear answers to these:

• What may our AI touch? The specific topics and tasks it's allowed to handle (hours, FAQs, bookings, etc.).
• What must it never do or say? Your no-go list from lessons 3 and 7 — no inventing prices, no refunds, no medical/legal advice, no moving money.
• When does a human step in? The handoff rule: anything about money, anything sensitive, anything it's unsure of.
• How do we tell customers it's AI? Your disclosure line from lesson 2.
• What data do we collect, and how long do we keep it? Your lesson-5 habits, in writing.
• Who handles a data request? The person and the steps, so a "show me / delete my data" email (lesson 6) is routine.

That's your policy. Six honest answers. Keep it where your team can see it, and update it when things change. (This is your capstone — fill it in for your real business.)

The vendor checklist

When someone pitches you an AI tool, you now have the eyes to judge it. Ask:

1. Where does it get its answers? You want "from your approved knowledge" (grounded), not "from the open internet" (improvising). — lesson 8
2. What does it refuse to do? A good vendor proudly names the dangerous actions it keeps locked off. "It does everything" is a red flag. — lesson 7
3. Does it disclose it's AI? It should, plainly and up front. — lesson 2
4. What happens to customer data — and can I see it and delete it? You need to be able to honor data requests. — lessons 5 & 6
5. How does it hand off to a human? There should be a graceful exit, not a bot that bluffs forever. — lesson 7

A tool that answers these well isn't just convenient — it's compliant by design, which means it keeps you safe without you having to babysit it.

Where this points

Everything in this course describes one kind of AI: a scoped, grounded, honest assistant that knows its limits and hands off to a person — the opposite of a free-roaming chatbot pulled off the open web. That's not an accident; it's what a responsibly built business AI looks like.

This is exactly the standard Portmint builds to. A Portmint assistant wears your business's name, answers only from your approved knowledge, keeps dangerous actions off by default, discloses that it's AI, and hands off to your team at the edges — the whole checklist above, baked in. If this course made you want AI you can deploy with a clean conscience, that's the work we do. And if you want to go deeper on building it well, the lighthouse keeps more courses lit at /lighthouse/courses.

Your turn

Right now, draft your one-page policy — even rough, even in a notes app. Six headings, your honest answers. The business that has thought through these questions before turning on AI is miles ahead of the one that figures it out after a complaint. You've done the thinking. Now write it down.

That's the voyage. You set out worried about getting into trouble, and you're leaving with a map, a policy, and a checklist. Use AI boldly — just use it like someone who knows where the rocks are. Fair winds. 🐙