A good AI receptionist is a useful tool. A great AI receptionist is a tool that knows when to hand the call to a person. This page is about the second one. It is the honest inventory of where the technology stops, where a human should take over, and how the handoff actually works in a garage door business.
If you are evaluating an AI receptionist and the vendor's pitch doesn't include the word "limitations," the pitch is incomplete. Every tool has a range. The question is whether the range covers the calls your shop actually gets, and whether the system is honest about the ones it doesn't.
There are two failure modes for an AI receptionist in a small shop.
The first failure mode is missing calls the technology should have caught. This is the obvious one. A caller with a broken spring at 7 PM hits voicemail instead of an AI, and the call goes to the next company on the list. The owner's frustration is real and the revenue loss is real.
The second failure mode is catching calls the technology shouldn't have caught. This is the less obvious one. A caller describing an injury, an irate customer demanding to talk to a manager, a confused elderly caller who needs patience the system can't deliver — these are calls that an AI handles badly, and the bad handling makes the situation worse. The shop that puts an AI on every call without an honest escalation path will, occasionally, turn a recoverable situation into a lost customer.
The whole point of an honest limitations list is to make sure your shop has a plan for both failure modes. The technology handles the calls it should handle. A person handles the calls only a person should handle. The handoff between them is clean.
The rest of this page is that list, with the calls named, the handoff explained, and the path to setting it up.
A short list. The full version is in calls an AI should never handle alone. The short version, in priority order:
1. Injury or safety calls. "The spring snapped and hit my hand." "My kid's fingers are caught." "I think my father fell in the garage." The right answer is 911, then a real person from your shop. The AI can capture the details and escalate, but it should not be the voice on the line giving medical or safety guidance. Set the system up to flag any mention of injury, blood, "hurt," "stuck," or "emergency" and route immediately.
2. Active emergencies with a person in danger. A door has collapsed, a cable has snapped, something is hanging in a way that could hurt someone. The AI's job is to get a human on the line quickly, not to handle the call. The escalation path is the same as injury: flag, route, hand off.
3. Irate customers demanding a manager. A caller who has been waiting three days, who paid for a part that didn't fix the problem, who is ready to leave a one-star review: that call is human work. The AI can de-escalate the surface heat — acknowledge the frustration, capture the complaint, get the caller to a person — but it cannot repair the relationship. The full playbook for handling angry callers on the phone, including the de-escalation patterns that work, is in handling upset callers with AI. The triggers that should pull a person in are in when to escalate an angry caller to the owner.
4. Legal threats or formal complaints. "I'm calling my lawyer." "I'm reporting you to the BBB." "I want to file a claim." These are calls where anything the AI says can be quoted later. The right answer is a polite "I want to get you to the right person" and a hand off. Anything beyond acknowledgment is human work.
5. Calls where the caller explicitly asks for a human. "Can I talk to a real person?" "Is this a robot?" "I need to speak to someone." The right answer is yes, immediately, and gracefully. The mechanics of that handoff are in AI-to-human handoff: how it actually works.
6. Confused, distressed, or vulnerable callers. An elderly caller who can't remember why they called. A caller who is crying. A caller who is clearly having a hard time and needs the kind of patience a person is better at delivering. The AI can be polite, but the patience that the situation needs is a human quality. The customer-experience angle is in what callers notice first on the phone.
That's the list. Most shops will get a few of these a month. The system should be set up to handle them.
A second tier of calls that the technology can handle but where a thoughtful handoff is better. These are not "never go to AI" calls; they are "go to AI first, with a fast handoff path" calls.
The complex new-door quote. A caller asking for a price on a custom door with insulation, windows, and a specific color. The AI can capture the basic details and book an in-home estimate. A human could do the same thing, and a human could potentially give a rough range over the phone. The trade-off is real: the AI is more consistent at capturing the details, the human is better at handling the ambiguity of a custom quote. Most shops let the AI book the estimate and have the human do the quote on site.
The call that goes long. A caller who needs to talk through the door's history, the warranty situation, the insurance claim, the property management company. The AI can hold the conversation for a while. Past three to four minutes, the call is in the territory where a human does it better. A reasonable system is to let the AI handle the first three minutes, then offer a callback from a person with all the details captured.
The caller with a hearing or speech impairment. The current generation of voice AI assumes a voice on both ends. For TTY callers, the right answer is a different channel: a text number, an email, a web form. The system can be set up to offer those channels on the call, but a person on a relay service is often the better answer.
These edge cases are the place where a thoughtful setup makes the difference between a system that books jobs and a system that frustrates the right callers. The full inventory of where the line sits is in calls an AI should never handle alone, and the recovery pattern when the system gets confused is in what happens if the AI gets confused mid-call.
Three patterns, in order of how often they get used in a garage door shop.
1. Warm transfer. The AI answers, captures the basics, identifies that the call needs a human, and bridges the call to a person — the on-call tech, the owner, or a designated escalation number. The caller doesn't have to repeat their name, address, or issue; the AI passes the captured details to the human. This is the cleanest pattern. It works in a shop that has someone available to take the escalation call.
2. Same-call escalation with a callback promise. The AI doesn't have a person to bridge to in the moment. It captures the call, reassures the caller that someone from the shop will call back within a defined window (e.g., 15 minutes for emergencies, 2 hours for routine), and routes the escalation to the owner's phone with the details attached. This is the right pattern for a solo owner who's on a ladder. The callback race in the morning is replaced with a real promise and a real person.
3. Take a message and follow up. For the calls that aren't urgent but are off-pattern — a billing question, a warranty claim, a complex quote — the AI captures the details and the shop follows up in business hours. The full mechanics of this path are in AI-to-human handoff: how it actually works.
A shop that has all three patterns set up — warm transfer, callback promise, and take-a-message — covers almost every situation. The first is the most-used pattern, and the second is the most-used pattern for solo operators.
A caller at 4:15 PM on a Friday. Example, with one realistic call:
"I want to talk to the owner. I had a spring replaced two weeks ago and it broke again. I want my money back or I'm calling my lawyer."
The AI hears the right keywords: "owner," "money back," "lawyer." The right path is:
A system that tried to handle this call alone — apologizing, promising a refund, negotiating — would create a mess. The owner would then spend twice as long untangling what the AI said. A system that hands off cleanly is a system the owner can trust.
The reliability question for an AI receptionist is not "does it ever fail." It is "what happens when it does." A reliable system has three properties:
The full reliability question is in is voice AI reliable enough for my business line. The point for a shop owner is that reliability is not "always works." It is "fails well."
The cost of having a system that says "I'll get you to a person" is a person. If the shop is solo and the owner is on a ladder, the callback is 30 minutes away, not 30 seconds. The customer experience is different. That is a real cost and it is worth naming.
The honest counter is that the alternative — leaving every call to voicemail, or routing every call to the owner's cell during a job — has a worse customer experience on the calls the AI would have handled. The total is the right comparison: how many callers get a clean answer from the AI, how many get a clean handoff to the owner, and how many fall through. The fall-through number is what the limitations list is meant to drive to zero.
The right setup is one where:
If your numbers are different from that, the setup needs adjustment. The first thing to check is whether the escalation rules are written for your shop's reality. The full case for a thoughtful setup is in calls an AI should never handle alone and what happens if the AI gets confused mid-call.
AI receptionist limitations are real, nameable, and small in number. The calls that should always go to a human are injury, active emergencies, irate customers, legal threats, explicit requests for a human, and confused or vulnerable callers. That's a few a month for most shops. The handoff to a person is a real pattern with three flavors — warm transfer, callback promise, take a message — and the right setup covers all three.
The shop that puts an AI on every call without an honest limitations list will eventually mishandle one of these calls. The shop that builds the system with a real escalation path will handle them better than they handle them today. The technology does most of the work. The design does the rest.
The next step is to test the technology against the calls your shop actually gets. $97 first month, then $297/month flat, unlimited calls, no contract. Call the demo, throw a hard call at it, and see how it hands off.
Call the live demo and have Ava call you now — hear exactly what your customers will hear when they call your shop.