AI for Electronics Repair Shops: Intake, Status and Warranty

AI for Electronics Repair Shops: Intake, Status and Warranty

To use AI in an electronics repair shop, and especially a phone repair shop, start with what repeats every day: collecting the details of the fault, reporting repair status, and answering warranty questions from the shop's written policy. The technician still decides the diagnosis, the quote and whether warranty applies. The AI prepares the intake form, checks the work order and hands off anything that needs judgment. That cuts repeated questions on WhatsApp and gives time back at the bench.

Key takeaways

• Split customer service into three flows: defect intake, repair status and warranty questions. Each one has its own rules.
• The AI collects information and organizes the form. Diagnosis, quote amount and warranty approval stay with the technician.
• Status must come only from the work order. If the system does not have the information, the AI says it will check with the team.
• A device passcode, photos and conversations are effectively sensitive personal data. Define what can be requested and where it is stored.
• Start with a single flow, log every conversation and agree on a success criterion before expanding.

Where AI helps and where the technician decides

A repair shop receives, in essence, three kinds of message: someone who has just broken a device and wants to know what to do, someone who has already dropped a device off and wants an update, and someone who had a repair done before and now has a problem. All three are repetitive in form and delicate in content. The form can be automated. The content that creates liability should not be.

A simple division of roles helps you decide what goes into the project:

• The AI can: ask about symptoms, request photos, summarize the conversation, check status in the work order, explain the written warranty policy, share the shop's opening hours and pass the customer to a person.
• The technician decides: the likely cause of the fault, whether the device needs to be opened, the quote, parts lead time, whether a new fault is related to the earlier repair and whether warranty applies.
• Nobody should let the AI: promise a fix, state that data was or was not lost, deny a warranty claim, give a discount or say the problem is the customer's fault.

This is the same line that shows up in other businesses that deal with quotes and after-sales. It is worth comparing with AI for Auto Repair Shops and Dealerships: Quotes, Parts, Scheduling, where the logic of separating collection from decision is identical.

How to run defect intake on WhatsApp

Intake is the flow with the best return, because technicians lose a lot of time asking the same questions before they even look at the device. The goal is not to find the fault. It is to reach the bench with a complete form and no guesswork.

A question script the AI can run in plain language:

• What is the brand and model of the device?
• What exactly is happening? (won't turn on, won't charge, cracked screen, screen with stains, battery drains fast, no sound, no signal, overheats)
• When did it start, and was it sudden or gradual?
• Was there a drop, a knock, contact with liquid or charging with a non-original accessory?
• Has the device been opened or repaired before, here or elsewhere?
• Does it turn on and can it be used? If so, can the customer back it up before dropping it off?
• Which accessories will be left (case, SIM card, memory card, charger)?
• Preferred way to be contacted, and whether the customer agrees to receive updates on WhatsApp.

The AI turns the answers into a short form, with the customer's account in their own words and a list of what remains unanswered. The technician reads it in seconds and already knows what to confirm when the device is checked in.

Photos help, but they have limits. A photo of a cracked screen confirms the obvious, and a photo of a device that won't turn on says little. Use images as attachments to the form, not as the basis for a guess. The precautions around misreading are covered in AI Image Understanding for WhatsApp Photos: Uses, Errors, Limits.

One security detail: the AI should never ask for the passcode or unlock code in chat. If the bench needs it for testing, that is agreed in person and recorded on the drop-off form, with the customer's consent.

Quotes: what the AI can and cannot say

The most common mistake is treating the quote as a catalog question. In a repair shop, the same symptom can come from a simple part or from a board failure, and the difference only shows once the device is open. So the rule is direct: the AI does not give a repair price.

What it can usefully do at this stage:

• Explain how quoting works in the shop: the device is assessed at the bench, the technician sends the price and lead time, and work only starts after the customer approves.
• State, if the policy exists, whether there is an assessment fee and in which situations, using text approved by the owner.
• Warn that the quote may change if another fault is found, and that this will be communicated before any additional charge.
• Record the customer's approval or refusal with the date and the original message, and pass it to the technician.

The technician builds the quote with diagnosis, part, labor and lead time. The AI can help draft the message in clear language from the fields the technician filled in, but sending goes through human review for the first months. Customer approval must always be explicit, never assumed from silence.

How to report repair status without making things up

Asking how the device is doing is the most frequent message after drop-off. The risk is the AI answering with optimism about something nobody has verified. The fix is to tie the service to a single source: the work order.

Define a closed list of statuses that works for the team and the customer:

• Received, awaiting assessment.
• Under assessment at the bench.
• Awaiting quote approval.
• Awaiting part.
• Being repaired.
• Final testing.
• Ready for pickup.
• Delivered.

For each status, write a standard sentence the AI may use, with no invented timeframe. If the part's lead time is not in the system, the answer is that the team will confirm an estimate, and the conversation is flagged for a human. If the customer asks why it is taking so long, the AI reports the current status and escalates, without creating explanations.

Proactive notices, such as ready for pickup, work well but require consent and approved templates. The steps are in WhatsApp Outbound Messages: Templates, Opt-In and Best Practices. Also agree on the time of day notices go out, so customers do not get a message in the middle of the night.

Warranty questions: answer the policy, don't decide the case

Warranty is where a conversation turns into a dispute most easily. So the design has to separate two things: explaining the rule and deciding whether a case falls under it. The first can be the AI's job. The second always belongs to a person.

For the first, the shop needs a written policy, short and reviewed by someone who understands consumer law, taking into account what Brazil's Consumer Defense Code already provides. The AI answers only from that text. If a question is not covered, it does not improvise.

A decision guide for the warranty flow:

• The customer asks how the warranty works: the AI explains the written policy and says what to bring (the device, proof of purchase or the work order number).
• The customer says the fault came back: the AI finds the earlier work order, collects the account and photos, opens a return ticket and passes it to the technician. It does not say whether the case is covered.
• The fault seems to be a different one, unrelated to the repair: the AI treats it as a new intake, without offering an opinion on coverage.
• The customer is angry or threatens to complain publicly: the AI acknowledges the message, does not argue the merits and passes it to a manager immediately.
• The customer mentions signs of misuse, such as liquid or a drop: the AI records the account verbatim and leaves the assessment to the technician.

In practice, the technician examines the device, compares it with what was done before and communicates the decision. If the answer is no, the explanation should come from a person, with the technical reason. Decisions that affect the consumer call for a path to human review, a topic detailed in LGPD Automated Decision-Making: The Right to Review (Art. 20).

Customer and device data: LGPD precautions

A phone holds messages, photos, contacts and access to accounts. Even without opening anything, the shop handles names, phone numbers, models, IMEI numbers and accounts that may contain intimate information. Some precautions apply from day one:

• Ask only for what intake needs. IMEI and work order number, yes. Passwords, verification codes and banking details, never.
• Tell the customer to back up and remove accounts before drop-off when possible, instead of the shop asking for access.
• Define how long conversations and photos are kept and who can access them.
• Tell the customer that part of the service is handled by AI and that they can speak to a person.
• Check where the AI provider processes the data. If another country is involved, read International Data Transfer Under LGPD When Using AI Providers.

For official guidance on personal data processing in Brazil, the reference is the website of the ANPD, the national data protection authority. And because the agent reads messages from third parties, it is worth knowing the risk of malicious instructions hidden in text or images, described in OWASP's guidance on prompt injection. The practical rule is that the AI performs only planned, limited actions, such as checking status and opening a ticket, and never changes amounts or grants warranty.

Illustrative scenario and how to run a pilot

Illustrative scenario: a customer writes that their phone fell and the screen has a green stripe. The AI asks for the model, whether the device turns on, whether it has been opened and whether it touched liquid. It builds the form, attaches the photo and says the quote only comes after the technician's assessment. The next day, the customer asks how it is going. The AI checks the work order, sees it is awaiting quote approval and replies that the technician has already sent the proposal, repeating what the system says. Days later, the customer returns saying another stain has appeared. The AI opens a return ticket, attaches the history and calls the technician, without saying whether it is covered.

To put this into operation, the method we use has four moves:

• Map: record how the shop handles customers today, which questions repeat, where information lives and how much of the technician's time goes to messages. This is the baseline for comparison.
• Limit: choose the permitted data and actions. The AI reads the work order, writes the form and opens a ticket. It does not calculate prices, decide warranty or access the device.
• Pilot: start with one flow, usually intake or status, logging every conversation and with a success criterion defined beforehand, such as forms that reach the technician complete and fewer rounds of repeated questions.
• Transfer: at the end, the team reviews the logs, adjusts the answers and decides whether to expand, fix or end the pilot.

To measure the pilot, the ideas in How to Measure an AI Automation Pilot: Metrics and Decision Gates apply directly. The NIST risk management framework is also a good reference for organizing responsibilities and reviews.

Frequently asked questions

Can the AI give the repair price on WhatsApp? It is not recommended. The price depends on a diagnosis made with the device in hand, and a number given earlier tends to become a promise. The AI can explain how quoting works and collect the details so the technician can prepare the proposal.

Can the AI say whether a fault is covered by warranty? It can explain the shop's written policy and open a return request with the history of the earlier repair. Deciding whether the case is covered requires a technical assessment of the device and is a person's responsibility. This avoids wrongful denials and disputes with the consumer.

How do I stop the AI from reporting the wrong status? Tie its answers to the work order, with a closed list of statuses and standard sentences. If the information does not exist in the system, the AI says it will confirm with the team and flags the conversation. Reviewing a sample of conversations every week helps catch failures early.

Do I need a work order system to get started? It helps a lot, because it is the source of truth for status and history. If the shop still uses a spreadsheet or a notebook, you can begin with intake, which depends only on the conversation, and organize the work orders in parallel before automating status.

Sources

• ANPD: Autoridade Nacional de Proteção de Dados
• OWASP: prompt injection prevention
• NIST AI Risk Management Framework
• Anthropic: Building effective agents