Calls an AI should not handle alone come down to a short list: anyone hurt or in danger, anyone talking lawyers or lawsuits, and anyone too angry to be reasoned with by a booking flow. Everything else — bookings, triage, quotes on standard work, messages — is fair game for the AI. Those three categories need a human, fast.
This isn't a knock on the technology. It's the same judgment call you'd make with a brand-new CSR on their first day: some calls you hand to a trainee, some you take yourself. The difference is the AI follows the rule every single time.
1. Injuries and safety emergencies. A spring snapped and hit someone. A door came down on a person. A customer smells gas near an opener install. These calls need a human immediately — partly for the caller's sake, partly because everything said on that call may matter later. The AI's job here is to recognize the emergency language instantly and get a person on the line, or get the caller pointed at emergency services. Not to troubleshoot.
2. Legal threats. "You'll be hearing from my attorney." "I'm filing a complaint." "I want your insurance information." The moment a call turns legal, every word is potential evidence. A human needs to take it — calmly, minimally, and with a record. The AI should capture the caller's details and escalate, not attempt to smooth it over.
3. Irate escalations. There's a temperature past which a caller isn't booking anything — they want to be heard by someone with authority. Keeping them in an automated flow, even a good one, pours fuel on it. The right move is a fast, respectful escalation with a callback promise a person will keep.
Around those three, there's a gray zone worth naming: warranty disputes on finished work, big commercial accounts with contract questions, media or insurance adjusters calling about a claim. Not emergencies, but all judgment calls a person should own.
The pillar on AI receptionist limitations maps the full boundary between AI work and human work. This article is the hard edge of that map.
Owners hear "AI answers your phones" and picture it winging its way through every scenario. That's backwards. A good setup is a small exclusion list wrapped around a huge coverage area.
Do the math on your own call log. Say your shop takes 40 calls a week. Booking requests, "how much for a spring," "my opener's dead," reschedules, "are you open Saturday" — that's 35 of them, easy. The injury, the legal threat, the raging customer — maybe one or two a month, if that.
So the right question isn't "can the AI handle everything?" It's "does the AI catch everything and route the one-in-a-hundred call to me correctly?" A system that answers 100% of calls and escalates 2% of them beats a human who answers 60% of calls — because the 40% that hit voicemail included emergencies too, and voicemail doesn't triage.
Say a property manager calls about a door your tech serviced last week — the tenant says it's still not right, and now there's a question about whether your warranty covers the second visit. Nobody's hurt. Nobody's shouting. But there's money and a relationship on the line.
A well-configured AI doesn't try to adjudicate warranty coverage. It captures the property, the issue, the callback number, flags it for a human, and you call back with the job history in front of you. The caller gets a fast, professional response; you get to make the judgment call with full context. That's the handoff working as designed — the mechanics of it are in how AI-to-human handoff actually works.
Your list should match your shop, but here's a working starting point:
Write yours down before you go live. Then, once a month, scan the calls that escalated and ask whether the rules caught the right ones. If the AI is escalating routine price shoppers, tighten the rules. If a genuinely furious customer sat in the booking flow, loosen them. What happens on the calls where the AI itself loses the thread is a related but separate question — what happens if the AI gets confused mid-call covers that recovery behavior.
Calls an AI should not handle are a tiny fraction of your phone traffic — but they're the fraction that can cost you a customer, a review, or a lawsuit. The goal was never to remove humans from your phone line. It's to make sure humans only touch the calls where a human actually matters, and that none of those calls die in a voicemail box.
Define your short list, test it in the first week, and let the AI carry the rest.
Call the live demo and have Ava call you now — hear exactly what your customers will hear when they call your shop.