Home/Blog/Strategy
Strategy

Writing an AI policy for your support team (template included)

Most support teams enable AI without ever writing down what the AI can and cannot do. Three months later, an AI does something dumb and there is no one to blame except yourself.

SG
Stefan Goluwa
Founder · March 16, 2026 · 8 min read

I have seen this play out three times now. Team enables AI. AI does most of the work well for two months. AI does one stupid thing (refunds a charge that should not have been refunded, gives medical advice when it should not have, names a competitor in a product comparison). Team panics. Founder writes a policy in 20 minutes. Policy is bad because it was written under pressure.

Better to write the policy now. It takes an hour. Here is the template we use internally and recommend to customers.

The four-section template

Section 1: What the AI is allowed to do without human approval. Section 2: What the AI is allowed to do with after-the-fact review. Section 3: What requires a human in the loop before the action. Section 4: What the AI is never allowed to do.

Section 1: autonomous actions

Default: answer questions grounded in the KB, summarize conversations, draft replies for agents, classify intent and sentiment, propose tags, transcribe and summarize calls. The line: actions that do not commit the company to anything.

Section 2: autonomous with review

Default: refunds under $30 per order with 1 per customer per 30 days, store credits under $50, free shipping upgrades on existing orders, sending CSAT surveys, auto-translating replies. The line: small reversible commitments.

Section 3: human in the loop

If the action involves committing the company to something, the AI's job is to draft, not to send. Sending is the human's button.

Default: refunds over the cap, cancellations, plan changes, contract questions, hiring questions, anything from an account tagged VIP, anything from a journalist or a regulator. The line: actions that are not reversible or are public-facing.

Section 4: never

Default: legal advice, medical advice, financial advice, hiring decisions, signing contracts, agreeing to terms, naming a competitor, making promises about your roadmap, threatening or abusive language. The line: anything where the legal or PR exposure of getting it wrong is catastrophic.

The template at the end of this post

Plain-text AI Policy template at /resources/ai-policy-template.md. Copy, paste, fill in the blanks. Takes one hour. Save your team three months of cleanup.

#AI#policy
SG
Written by Stefan Goluwa · Founder
I founded message.com to replace four bad tools with one good one. I write about the wins, the misses, and the math.

Try Message free for 14 days.

Unlimited agents, flat monthly price. No credit card to start.

Start free trial