AI for Auto Repair Shops and Dealerships: Quotes, Parts, Scheduling

AI for Auto Repair Shops and Dealerships: Quotes, Parts, Scheduling

AI for auto repair shops and dealerships works best when it takes over the organizing work: collecting the details of a quote request, reporting service status and proposing appointment times. Diagnosis, price and any promised completion date stay with a person at the shop. The AI prepares the information, and the mechanic or service advisor decides and answers for it.

Key takeaways

• The AI should receive the request, ask for what is missing and hand the service advisor a ready summary, not a price or a diagnosis.
• Service status is the most repeated message of the day and the safest to automate, as long as it comes from the shop management system and not from the model's memory.
• Parts and scheduling require lookups in real data. Without them, the agent invents availability and lead times.
• Diagnosis, price, warranty and authorization of additional work require a recorded human approval.
• Start with one workflow, log every conversation, agree on a metric beforehand, and only then expand.

Where AI helps and where it should not decide

A repair shop or dealership service department runs on interruptions. The phone rings while the service advisor explains a job at the counter, a customer asks on WhatsApp whether the car is off the lift yet, and the parts clerk hunts for a part while someone requests a quote by voice message. AI is useful for absorbing that queue of repetitive messages and structuring what comes in. It becomes dangerous when it starts offering opinions on things only someone who has seen the vehicle can state.

A simple decision guide:

• Can be automated: collecting plate, model, year, the customer's complaint and photos; confirming an available time slot; reporting the service stage recorded in the system; reminding the customer about a scheduled service or a pickup.
• Automate with human review: a summary of the complaint for the quote; a draft message about an unavailable part; a suggested service type based on the vehicle's history.
• Do not delegate: fault diagnosis, final price, discounts, a completion date promised without the shop's confirmation, warranty, admitting fault or taking responsibility for damage.

The test is the consequence of an error. If a wrong answer could lead to a car taken apart without authorization, a disputed bill or a safety risk, the decision does not belong to the agent. The article on what not to delegate to AI in business goes deeper into this reasoning for other areas.

How to organize quote requests with AI

The quote is where a shop loses the most time before it earns any money. The customer writes something like "my car is making a noise in the front, how much will it cost?" Without plate, year, trim and a better description, nobody can answer, and the advisor spends the first exchange of messages just asking questions.

The AI's job here is to run that intake and return a structured record. A useful record usually contains:

• Plate and, when needed, VIN, or at least make, model, year and engine version.
• The complaint in the customer's own words, without technical translation added by the agent.
• When the problem shows up: cold start, at speed, when braking, in turns.
• Photos or a short video of the dashboard, the component or the noise, when the customer can send them.
• Approximate mileage and the date of the last known service.
• Contact preference and a time window for the callback.

The agent can ask for each item conversationally, without turning the exchange into a form. When the customer sends a voice message, the transcript helps the advisor, but it should be marked as an automatic transcript and checked. Transcribing WhatsApp voice messages with AI covers accuracy and privacy on this point.

The reply to the customer matters just as much. The agent can say that the technical team will assess the request and come back with a price and timeline, and it can explain what happens next, for example that some jobs require the vehicle to be at the shop for an inspection. It should not say the problem is probably the fuel pump, or that the repair will cost a certain amount. Even an informal estimate from the agent becomes an expectation, and the customer will hold the shop to it later.

For standardized services, such as interval-based maintenance, a different rule makes sense: if the shop keeps an approved price table, the agent can look up that table and present the amount, with the caveat that it depends on confirming the vehicle's condition. A person owns the table, and it stays outside the prompt text so someone can update it without touching the agent.

How to report service status without making things up

Questions about progress take up a large share of customer contact: is it done, are you waiting on a part, can I pick it up today. The answer already exists in the shop management system, on the work order board or in the spreadsheet the team uses. The agent needs to read that source, and only that source.

A good design defines a few states and translates each one for the customer:

• Received and awaiting assessment.
• Quote sent, awaiting customer approval.
• Approved, waiting for a part or an open bay.
• In progress.
• In final inspection or road test.
• Ready for pickup.

Each state has a message approved by the shop. If the system has no up-to-date information, the agent says it will check with the team and opens a task for the service advisor instead of guessing. This is a design rule: when the data does not exist, the correct response is to hand off.

For outbound messages, such as telling a customer the car is ready, WhatsApp requires consent and approved message templates. Read WhatsApp outbound messages: templates, opt-in and best practices before configuring any automatic notification.

Parts: check availability without promising what is not confirmed

At repair shops and dealerships, the bottleneck is often the part. The agent can help in two ways: preparing the list of likely parts for the parts clerk to verify, and relaying to the customer the situation the team has already confirmed.

The typical mistake is letting the model answer compatibility questions from memory. Part numbers, variations by vehicle trim and substitutions between manufacturers are data that must come from the catalog or the shop's system. If there is no catalog integration, the agent only records the need and passes it to a person. If there is one, it queries in read-only mode, and the result goes back as information for the parts clerk to validate, not as a commitment to the customer.

The control point is lead time. Stock on hand is a fact; supplier arrival time is an estimate. The agent can repeat what the responsible person recorded, along with the date it was recorded, but should not project on its own. If the customer asks when the part arrives and the record is old, the answer is to route the question to purchasing.

For shops that use an ERP or management system with an API, integrating AI with Brazilian ERPs such as Bling, Omie and Tiny shows the options and the risks of connecting an agent to a system that holds inventory and finance data. The rule of starting with read-only access applies here too.

Scheduling: propose times without wrecking the bay schedule

Scheduling a repair is not like scheduling a medical appointment. Each job takes a certain type of bay, a lift, a mechanic with a certain specialty and, sometimes, a part that has to arrive first. If the agent offers any free slot on the calendar, it creates bookings the shop cannot honor.

The practical path is to turn the shop's rules into data the agent can consult:

• Service types with a standard duration and the resource required (bay, alignment, electrical, body work).
• Drop-off and pickup windows.
• A daily limit on long jobs.
• A rule for jobs that depend on a part: schedule the work only after the part is confirmed.
• A rule for cases that need an inspection first: the booking is for an assessment, not for the repair.

With that in place, the agent offers options within what is feasible, confirms with the customer and writes the booking to the system. If the request falls outside the rules, such as an urgent case, a car stranded on the road or an unusual job, it passes the case to the advisor. The article on automatic appointment scheduling on WhatsApp with AI describes how it works technically. The logic is the same, except that at a repair shop the resource is the bay and the mechanic, not just a time slot.

Reminders for booked dates reduce no-shows, and the agent can also offer rescheduling when the customer says they cannot make it, which frees the slot for someone else.

Illustrative scenario: from request to approved quote

The following scenario is illustrative and does not describe any real shop.

A customer writes in the evening saying her car pulls to one side when braking. The agent asks for the plate, year and trim, asks whether the problem happens at any speed and whether she can send a short video. She sends the plate and a description. The agent builds the record, links the vehicle to the shop's history, sees that the brake pads were replaced earlier and notes that as information without drawing a conclusion. It replies that the technical team will review the request on the next business day and that, because this involves the brakes, the shop recommends inspecting the vehicle in person, offering two available inspection slots.

The next morning, the service advisor finds the record ready, confirms the time slot and, after inspecting the car, determines the diagnosis and the price. The quote is sent by the advisor, or through a message the advisor approves. When the customer approves, the agent records the authorization with date and time, and the work order moves forward. During the repair, the agent answers status questions from the system. If the mechanic finds an additional problem, the agent does not ask for authorization on its own: the message comes from the advisor, because it involves price and risk.

What this workflow preserves: no diagnosis came from the agent, no price was stated without approval, and every step was recorded.

How to start: map, limit, pilot, hand over

The method we apply to this kind of project follows four moves.

• Map: describe how quote requests, status questions and bookings flow today, who answers, how long it takes and where customers are left without a reply. The guide on mapping business processes before applying AI works as a roadmap. The result is a baseline: without one, nobody knows later whether anything improved.
• Limit: define which data the agent can access (for example, work orders in read-only mode, the calendar, the table of standardized services) and which actions it can take (create a record, propose a time, log a task). Price, diagnosis and authorization stay out. Limits should also exist on the output side, and guardrails for AI agents shows how to validate what the agent says before it reaches the customer.
• Pilot: pick one real workflow, for example quote request intake, log every conversation and agree on a metric beforehand, such as time to the first useful reply or complete records per request. How to measure an AI automation pilot details the decision criteria.
• Hand over: give the team the documented workflow, with who reviews the logs, who updates the tables and what to do when the agent gets something wrong. Then decide whether to expand, adjust or stop.

In managed operation, Be Human (be-human.ai) handles monitoring and maintenance, but the business rules, the pricing and the diagnosis remain the shop's.

Personal data, photos and responsibility

A repair shop holds names, phone numbers, addresses, plates, VINs and vehicle usage history. A plate and a VIN can identify a person, so they are treated as personal data in combination with the rest. Photos and videos sent by customers may show faces, documents or other vehicles in the background.

Minimum good practices: collect only what the service requires, tell the customer the conversation is supported by AI, define how long conversations are kept and restrict who can read those records. The company decides and answers for this data. In Brazil, the ANPD (the national data protection authority) is the official source of guidance, and the LGPD and artificial intelligence guide covers what companies should do. For controlling the agent's access to systems, the OWASP Authorization Cheat Sheet helps apply least privilege. And on the design decision itself, Anthropic's article "Building effective agents" explains why to start with simple, predictable workflows before giving an agent autonomy.

Frequently asked questions

Can the AI give the customer a quote directly? Only for standardized services, using a price table approved by the shop and with a caveat that it needs confirmation. For any service that depends on inspecting the vehicle, a person must set and approve the price before it reaches the customer.

Can the AI diagnose a problem from an audio message or photo? It can organize what the customer described and highlight useful details, but it should not conclude what causes the problem. Diagnosis requires inspection, technical knowledge and accountability, and that stays with the mechanic or service advisor.

Do dealerships and small repair shops use the same design? The logic is the same: structure the request, read the status and propose a schedule within rules. What changes is the complexity of the integrations, because a dealership usually has more systems and more approval steps. At a small shop, the starting point can be a well-kept spreadsheet and WhatsApp.

Where do we start without putting customer service at risk? Choose a single workflow, preferably quote request intake or status inquiries, and keep a person reviewing the conversations. Define beforehand what success looks like and what would make the pilot stop. If the team gains clarity and customers get faster replies, expand step by step.

Sources

• Anthropic: Building effective agents
• OWASP: Authorization Cheat Sheet
• ANPD: Autoridade Nacional de Proteção de Dados
• NIST AI Risk Management Framework