Using AI Without Losing Customer Trust: Where to Draw the Line
Which customer conversations can AI handle, and which need a person? Where to draw the line, plus five questions to ask before you launch.
AI & HUMAN PERFORMANCEAI IMPLEMENTATION & STRATEGY
9/10/20262 min read
When AI talks to your customers, it speaks as you. A tribunal made that point in 2024, when it held Air Canada responsible for wrong information its website chatbot gave a traveler. The airline argued the chatbot was a separate entity, but the tribunal didn't buy it.
While this is an extreme case, it points at the right question for every business. Not whether to use AI with customers, but where to use it and who's checking it.
Start with a gut check
Your customers trust you, not your software. So ask yourself: can I stand confidently behind everything AI sends out under my name? Every message, from a one-line reply to a full proposal.
If the answer is yes, you're in good shape. If it's no, or even "mostly," that's your signal. Either a person checks that work before it goes out or regularly afterward, or it shouldn't go to AI at all.
Match human involvement to the risk
AI makes mistakes. Some customer touchpoints can absorb that, and some can't. The higher the stakes, the more a person should be involved. You can use the three tiers below as a rule of thumb (not legal advice), from most human involvement to least.
Tier 1: Keep it with a person
Anything that makes a commitment. Prices, quotes, refunds, guarantees, deadlines, exceptions to policy. You can be held to whatever gets said.
Complaints, bad news, and billing disputes. Upset people want to know a person is listening.
Anything that sounds like legal, tax, medical, or financial advice.
First personal contact with someone a client referred. Referrals are your best asset. Don't hand them to a script.
Tier 2: Let AI draft, but approve before it goes out
Emails, routine follow-ups, meeting summaries, first drafts of proposals. The AI saves you the typing, but a person owns what goes out.
Tier 3: Let AI answer, then review afterward
Approved answers only: hours, service area, what to bring, how booking works, etc.
After-hours message capture that a person follows up on the next morning.
Auto-scheduling inside clear limits.
This is the only tier where AI talks to your customers directly, so make sure someone is reviewing samples of these conversations at least weekly for continued accuracy.
Five questions before you launch anything customer-facing
What's the worst thing it could say, and who would that hurt?
Could we undo it? A wrong reminder is fixable. A wrong quote may not be.
When it gets something wrong, who fixes it? And do they know it's their job?
How does a customer reach a person quickly?
If an AI mistake costs a customer money, who is liable? Assume it's you, and ask whether you could afford to make it right.
If you can't answer all five with confidence, that's where you draw the line. Don't let AI talk to customers directly on that touchpoint. Keep it in Tier 1 or Tier 2 until you can.
