AI for Home Service Providers: Quotes, Site Visits, Follow-Up

AI for Home Service Providers: Quotes, Site Visits, Follow-Up

AI for home service providers works best as an organizing assistant. It receives the request on WhatsApp, asks for the photos and details a technician would ask for anyway, proposes times for the site visit and checks in with the customer after the job. Pricing, diagnosis and warranty stay with a person. The gain is quotes built on complete information, fewer wasted visits and follow-up that does not depend on the owner's memory.

Key takeaways

• AI should collect information and organize the schedule. It should not diagnose or settle a price on its own.
• Photos, short videos and direct questions cut down on visits that exist only to find out what the problem is.
• Automatic scheduling needs clear rules: service area, time windows, urgency and when to hand off to a person.
• Post-service follow-up (confirmation, questions, warranty, review) is the most neglected part and the simplest to automate safely.
• Start with a single flow, with conversation logs and one agreed metric, before expanding.

Where AI really helps service businesses

Electricians, plumbers, cleaning crews, pest control, air-conditioning maintenance and similar trades share the same routine. The customer reaches out on WhatsApp, describes the problem vaguely, sends a voice note with loud background noise and wants to know the cost. The person answering is in the field with their hands full and replies when they can.

The bottleneck is rarely the work itself. It is the queue of small tasks around it: asking for the address, requesting a photo, understanding the size of the problem, finding a time slot, confirming the visit, reminding the customer, asking whether everything went well. Each one is simple and repetitive, which is exactly where AI fits.

What does not fit loose, unsupervised AI: stating the cause of an electrical or plumbing problem, promising a date without looking at the real schedule, quoting a fixed price without a calculation rule and deciding warranty questions alone. For a broader view of that boundary, read what not to delegate to AI in your business.

How to collect photos and details before the quote

A bad quote starts with bad information. If the customer only writes that the faucet is leaking, the professional has two options: guess or drive to the site. AI can run a short, specific conversation for each type of service, in the tone of someone who handles customers well.

The logic is a script per service type, with questions in the order the professional would ask them. The AI follows the script, adapts the wording and records the answers in a single summary for whoever will prepare the quote.

A collection checklist to use as a starting point:

• Address or neighborhood, to confirm it is inside the service area.
• Property type (house, apartment, commercial space) and whether access is easy or restricted by a doorman or building rules.
• A description of the problem in the customer's own words, without correcting it.
• A wide photo of the area and a close-up of the problem spot.
• A short video when the problem involves sound, running water or intermittent behavior.
• A photo of the label or model number of the equipment, when there is one.
• Since when it has been happening and whether anything changed beforehand (renovation, part replacement, power outage).
• Stated urgency and whether there is immediate risk, such as a burning smell, sparks or water near an outlet.
• Best time window for the visit and who will be on site.

For cleaning, the script changes to approximate square footage, number of rooms, floor type, pets, type of dirt (post-renovation, routine upkeep, move-out) and special items such as high windows or upholstery. The structure is the same: short questions, one at a time, with an example answer when the customer gets stuck.

If the customer prefers to send voice notes, the AI can transcribe and summarize them, as long as the summary is checked and the transcript is handled carefully. Details on accuracy and privacy are in transcribing WhatsApp voice messages with AI.

What to do when there is a safety risk

The script needs a safety exit. If the customer mentions a gas smell, a melted wire, water touching the electrical system or a leak that will not stop, the AI does not continue the questionnaire. It gives the basic safety step you defined (for example, switching off the breaker or closing the main valve, if that is safe and the customer knows how), says that a professional will be notified and hands the conversation to a person immediately. The professional defines what counts as an emergency, in writing, before the flow is switched on.

What AI can and cannot say about price

This is where most pilots stumble. The customer asks for the price right away, and the temptation is to let the AI answer.

There are three possible stances, from safest to riskiest:

• No price: the AI explains that the quote depends on an assessment, collects the information and tells the customer when the proposal will arrive.
• Range by rule: if the service is standardized (for example, installing a simple item), the AI gives the range you registered, making clear it is an estimate subject to the visit.
• Automatic quote: only makes sense for fixed-scope services, with a price table maintained by you and human review before sending.

At the start, the first stance is the most prudent. The AI prepares the quote request with photos, description and address, and the professional replies with the price. That alone removes a good share of the back and forth. Once the table is mature and the flow has been measured, you can move to ranges.

Whatever the stance, the AI does not invent prices, does not promise discounts, does not confirm an execution date without checking the schedule and does not say the problem is simple or serious. These rules go into the prompt and also into checks outside it, as explained in the article on guardrails for AI agents.

When the quote goes out and the customer goes quiet, the follow-up can be assisted too. The article on following up on unanswered quotes shows how to remind without nagging.

How to schedule the site visit without back and forth

Scheduling looks trivial until the details appear: travel between one customer and the next, jobs that run longer than planned, customers who are not home, a different professional for each type of work.

For the AI to schedule well, it needs structured data, not common sense. The minimum:

• The real calendar of each professional or crew, in a tool the AI can query (Google Calendar is the most common).
• Service windows by region, so it does not propose times that force a trip across the city.
• Estimated duration per visit type, including travel margin.
• An urgency rule: what gets same-day service and what waits for the next window.
• A rescheduling and cancellation rule, with what the AI can do alone and what needs approval.

The typical flow is to offer a few time options, confirm the address and the contact of whoever will receive the professional, record the appointment and send a confirmation that says what the customer should leave accessible (breaker panel, shutoff valve, service area). Close to the date, a reminder reduces no-shows. To see how this mechanism works under the hood, read automatic appointment scheduling on WhatsApp with AI.

Illustrative scenario

Imagine a small residential electrical maintenance company. A customer writes that the kitchen outlet gets hot and sometimes trips the breaker. The AI picks up the risk word (hot), asks whether there is a burning smell or sparks and advises the customer not to use that outlet until it has been assessed. It then asks for a photo of the outlet plate, a photo of the breaker panel and what is plugged in there. With the answers, it builds a summary for the electrician and flags the case as a priority. The electrician reviews the summary and decides a visit is needed, and the AI proposes two times within that neighborhood's window. The customer picks one, receives the confirmation and a note to keep the panel accessible. At no point did the AI say what the cause was or what it would cost.

Note the design: the AI organizes and prioritizes, the professional assesses and decides.

How to follow up after the service

Once the job is done, most providers disappear from the conversation. That is a waste, because it is the moment when the customer is most willing to reply and to recommend you.

A simple follow-up, with short and useful messages:

• Confirmation of what was done, using the summary the professional recorded, the agreed warranty and recommended care.
• An open question a few days later: did the problem come back, any questions left?
• Sorting of the replies: if the customer reports a problem, the AI opens a ticket with photos and passes it to a person. If everything is fine, it asks for a review.
• A reminder for periodic maintenance, when it makes sense for the service (filter cleaning, panel inspection, a new pest treatment).

AI also helps separate complaints, questions and requests for new work, so each one reaches the right place. A complaint with an irritated tone should go straight to a human, which ties in with the idea of handing off to a person when sentiment calls for it.

A note on messages that start with you rather than the customer: on WhatsApp, outbound messages sent through the API require consent and approved templates. Before building reminders and surveys, read about outbound messages, templates and opt-in.

Data, photos and LGPD: what providers often forget

Photos of a home or business show more than the problem: documents on a table, faces, storefronts, routines, sometimes children. Address and phone number are personal data, and Brazil's LGPD applies to the provider even when the business is small.

Basic practices that fit any operation:

• Ask only for what the service requires. If a photo of the breaker panel is enough, do not ask for a video of the whole house.
• Tell the customer in the first interaction that they are talking to an automated assistant and how their data will be used.
• Define how long photos and conversations are kept and who can access them.
• Keep customer data from being pasted into open tools without control. The team needs a clear rule on this.
• Know where the AI provider processes the data, which connects to the discussion about international transfer.

For official guidance, the website of Brazil's data protection authority (ANPD) is the reference. To understand the data transfer question, the article on international data transfer under LGPD when using AI providers details what to check.

There is also a technical risk: a customer can, on purpose or by accident, write instructions that try to change the assistant's behavior (for example, asking for a discount as if it were a rule). That is why the AI should only have access to the actions that are strictly necessary. The OWASP list of risks for LLM applications describes these problems clearly.

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

The approach that makes sense for a service provider follows four moves.

Map: describe how requests arrive today, who answers, where the information lives, where customers drop off and what the most frequent loss is (wasted visit, incomplete quote, customer not home, no follow-up). Write down the starting point before changing anything.

Limit: decide what the AI can read (conversation, photos, calendar), what it can do (ask, summarize, propose times) and what it never does (give a fixed price, diagnose, promise a warranty). Fewer permissions mean less risk.

Pilot: choose one flow, for example only the information collection for one type of service, with a log of every conversation and one agreed metric, such as the share of requests that reach the professional with complete photos and data. The way to measure it is in how to measure an AI automation pilot.

Hand over: deliver the operation with documented scripts, named owners and an explicit decision at the end of the pilot: expand, adjust or stop.

A step-by-step plan for the first weeks:

• Pick one frequent service and one channel (the WhatsApp your customers already use).
• Write the question script with the person who does the work, not only the person who answers messages.
• Define the emergency words and situations and the path to a human.
• Register service areas, time windows and visit durations.
• Review the first conversations every day, fixing the script as you go.
• Only then switch on reminders and post-service follow-up.
• At the end, compare with the starting point and decide the next step.

For fundamentals on structuring language-model systems in a simple and controllable way, Anthropic publishes a good piece called Building effective agents, which argues for starting with the simplest possible design.

Automate the repeated question, not the decision that carries responsibility.

Frequently asked questions

Can AI send a quote to the customer on WhatsApp? It can, but only if there is a calculation rule maintained by you and, at the start, human review before sending. For services that depend on an on-site assessment, the safest approach is for the AI to collect photos and details and leave the price to the professional.

Do I need an expensive system to start? Not necessarily. A pilot flow can use the WhatsApp you already have, a digital calendar and a well-written script. What matters is the design of the questions, the limits and the conversation log, not the number of tools.

Does AI replace the person who answers messages or the professional who makes the visit? No. It handles the repetitive part: asking, organizing, reminding and following up. Technical assessment, the final quote, the work itself and sensitive situations stay with people, and the customer should be able to reach one whenever they want.

What should I do with customers' photos and data? Ask only for what is needed, tell customers how it will be used, set a retention period and restrict access. Because this is personal data, LGPD applies, so it is worth consulting ANPD guidance and, if possible, a digital law professional before scaling the operation.

Sources

• ANPD: Autoridade Nacional de Proteção de Dados
• Anthropic: Building effective agents
• OWASP Top 10 for LLM Applications
• NIST AI Risk Management Framework