An AI front desk for auto repair shops
Wait or drop off is the question that decides the booking. The five vehicle details to capture before anybody picks up, the safety questions that must never get an answer, and why a repair price quoted by text becomes an argument at the counter.
Sohail Safdar · · 6 min read
The service advisor is at the counter with a customer, both hands busy, and the phone rings. The caller wants to know what an oil change costs.
That question does not need a person. "My car makes a grinding noise when I brake" does, and it also needs five pieces of information before anybody can usefully help. An AI front desk handles the first kind and prepares the second, and the preparation is the half worth more.
What it answers from your facts
- Prices for routine, fixed-scope services: oil changes, rotations, alignments, inspections, filters.
- Whether you work on a particular make, and whether you handle diesel, hybrid or EV work. This is a surprisingly high-volume question and a cheap one to get right.
- Hours, parking, loaner or shuttle policy, and how long common services take.
- Warranty and payment terms.
- Your diagnostic fee, if you charge one, and whether it comes off the repair.
Anything not on the list is logged as a question rather than guessed at.
The price line, and why it matters here
A routine service has a price. A repair does not, until somebody has looked.
A brake job depends on what the inspection finds, which parts are available for that VIN, and what else turns out to be worn once the wheel is off. A number given by text is a number your advisor has to walk back at the counter, with the customer already holding it and entirely reasonable about being annoyed.
So: publish routine prices, quote the diagnostic fee, and let the repair price come from somebody who has seen the car. The honest version reads better than the evasive one:
"A brake inspection is $45 and comes off the repair if you go ahead. What it costs depends on what we find, and we will call you with a price before doing anything."
That answers, sets an expectation, and commits to a call. "Prices vary, please call us" does none of the three.
The five details to capture before anybody picks up
- Year, make and model.
- Mileage, if they know it.
- What the problem is, in their own words. Not a summary. "Grinding when I brake, worse turning left" is diagnostic information; "brake issue" is not.
- Whether a warning light is on, and which one if they can say.
- Whether the vehicle is drivable.
Those five convert a callback into a conversation. The advisor rings back already knowing what they are dealing with, instead of starting from "so what is the car?".
Points four and five also route the enquiry: a car that is not drivable is a tow and a different conversation, not an appointment slot.
The safety questions that must never get an answer
This is the hard line in this trade and it is easy to miss, because it does not look clinical.
- "Are the brakes safe to drive on?"
- "There's smoke coming from the hood, can I bring it in?"
- "The steering's gone loose, is that OK for a few miles?"
- "It's overheating, should I keep going?"
- "There's a strong fuel smell."
Every one of those is a safety question about a vehicle nobody has inspected, and the honest answer requires seeing it. A reassuring reply to "is it OK to drive it to you" is the worst message this business can send, and it is exactly what an unconstrained assistant will produce, because the question is phrased as a request for permission.
Set them as escalation keywords in the words customers use: brakes, smoke, smoking, steering, overheating, overheat, fuel smell, gas smell, grinding, tow, won't start, warning light. A false alarm costs an advisor a minute.
And the handover reply must not editorialise. "I am getting an advisor for you now" is right. Anything that implies the car is fine, or that it is not, is an opinion the software may not have.
Wait or drop off, which decides the booking
The single question that determines which slot the job needs.
A thirty-minute oil change somebody waits for and a day-long brake job they drop off are different shapes in the diary. Asking late means either a customer sitting in reception for four hours or a bay held empty for a job that took twenty minutes.
Ask it early, and set each service with its real duration, including the time the car occupies a bay rather than the time a technician is working on it. Those are different numbers and the diary cares about the first one.
Previous work, and complaints
Any question about work you already did goes to a person. Always, and whether or not it sounds like a complaint.
A customer saying "you replaced that six months ago and it's doing it again" may be reporting a warranty claim, a misdiagnosis, or an unrelated fault. All three need somebody who can look up the job, and none of them is improved by an automated reply that gets the category wrong first.
What to measure
- Calls and messages answered without an advisor, which is the volume case.
- Enquiries that arrived with all five vehicle details captured. This is the one that tells you whether the capture is working, and it is the number that makes advisors like the system.
- Time from handover to an advisor's first message. Usually the real bottleneck, and nothing about the software fixes it.
- Safety escalations, read weekly, because the list is how you learn which phrasings your customers use.
In Frontiva
The assistant answers from your own service list, prices and policies, in your wording, and logs anything it does not know rather than inventing a figure.
The safety list above is handled by your own escalation keywords plus a sensitive-topic check that runs before the answering path, as a deterministic pattern check with no model in it, so a question phrased as a request for permission cannot talk its way to an answer. A handed-over conversation is marked as needing a person and carries the customer's own words rather than a summary, which matters here because the customer's phrasing of a noise is diagnostic information.
The five vehicle details can be collected as an intake form attached to the service, asked at booking, with the answers on the appointment rather than in a thread somebody has to read. Services carry their real duration and their buffers, so the booking page only offers times that actually fit, which is the wait-or-drop-off problem solved in the diary rather than in the conversation.
A diagnostic enquiry becomes a lead with an owner and a follow-up task, so the callback does not depend on somebody remembering it between two jobs. What an AI receptionist should refuse to answer covers where the line sits across trades.