At 11:20 a.m. the estimate goes out by text: front pads and rotors plus a brake fluid flush, $1,180. The customer is in a meeting. At 12:15, on her lunch break, she texts back: "what's the flush for, and is the warranty parts and labor?" The service advisor is under a car. The Guru answers from the shop's own service and warranty information in seconds. Then she asks, "is it safe to wait a month on the rotors?" That one is a judgment about a car only the advisor has seen, so it goes to him with her question attached, and he calls back at 12:40 already knowing what she asked. At 12:50 she texts again:
Customer: ok go ahead with all of it Guru: To confirm: you're approving estimate #4471 for your 2019 Camry: front brake pads and rotors, and a brake fluid flush, total $1,180 plus tax. Shall I record your approval? Customer: yes Guru: Done. Your approval is recorded on the estimate, and the shop will start the work this afternoon.
She approves before her lunch break ends. Without that, it's voicemail, phone tag and a car sitting in the bay until tomorrow.
The gap between estimate and yes
That's the estimate-to-approval gap, and it's a different loss than the missed call. The customer has already come in and already has her car in the shop's bay. This isn't about attracting her. It's about not losing her, or her afternoon, at the approval step.
Here's how it usually goes. The shop calls with the estimate, and the customer doesn't pick up because she's at work. The shop leaves a voicemail. She calls back at lunch, but now the advisor is with someone else, so she gets voicemail too. Most people don't listen to voicemail, so the approval that should take minutes takes hours or drags into the next day, and sometimes the frustrated customer just says forget it. Every hour the car sits waiting is an hour the bay isn't earning.
Why the waiting hurts
A car repair is stressful. It's often unexpected, usually more than the customer hoped to spend, and she's without her car. Phone tag lands on someone who is already anxious about the bill. A customer who can't get a question answered about her own estimate is losing trust in real time, at the exact moment the shop needs her to approve a significant bill.
Two-way, and split the right way
Approval isn't just the shop reaching the customer. It's the customer being able to text the shop back with questions while she's looking at the estimate. Those questions come in two kinds:
- General questions, what does this service involve, is the part under warranty, can I pay in installments. The Guru answers these instantly from the shop's own information.
- Questions about this car, is this really necessary, what happens if I wait. These are safety judgments about a vehicle only the advisor has seen, so they go to the advisor with the question already captured, and he calls back ready to answer.
Some approvals need the advisor's judgment and relationship from start to finish: the complex repair, the customer who needs to be walked through options. A good setup routes those to a person. The point is to stop the approvals that only needed a quick answer from stalling in phone tag.
Making the yes safe
The approval itself is the one step where a mistake costs real money, so it is not left to the AI's judgment. When the customer says "go ahead", a deterministic workflow engine we built, JSFE, takes over. JSFE connects to the systems the shop has credentials for, including its shop management system, and from that point:
- It finds the real estimate in the shop's system, for the customer who is texting, so it can only record approval on an estimate that exists and belongs to her.
- It offers only valid choices: the lines on that estimate. If she approves the pads and rotors but not the flush, it reads back only those lines and that total.
- It validates each answer. "Probably" is not a yes, so the step asks again or goes to the advisor.
- It reads back the exact lines and total from the shop's own data, not from the model's memory, and records the approval only on an explicit yes.
- Anything else, a question, a "maybe", a change, cancels the step and goes to the service advisor.
Those steps are 100% accurate: the worst case is a missed reply caught on the next message, never an approval the customer didn't give. General answers are grounded in the shop's own content, with 97%+ accuracy.
After hours
A lot of approval delay hides after hours. Many customers can't look at an estimate until they're off work, when the shop is closed. If they can ask their questions and approve that evening, the work is waiting to start when the shop opens, instead of the car sitting through one more round of phone tag. Questions that need the advisor go to him first thing, already written down.
Questions to ask about any approval process
- How long, on average, does a car wait between the estimate going out and the yes coming back?
- Can a customer ask a question about the estimate by text and get an answer right away?
- When the AI records an approval, does it read back the exact lines and total from the shop's own system first?
- When the customer is unsure, does the question reach the advisor with the context attached?
The estimate-to-approval gap costs shops time, bay space and trust, not because of anything wrong with the repair work but because reaching the customer is harder than it should be. Answer her questions while she's looking at the estimate, and the yes comes in minutes. At 12:50, on her lunch break, it did.
