AI for home service businesses: calls, dispatch and follow-up

Today, AI can reliably answer and qualify calls around the clock, book a window into your schedule, flag emergencies to whoever is on call and draft follow-ups. Pricing, safety decisions and dispatch should run on fixed rules or stay with a person. In Lane, the dispatch system traced below, the AI handles one step out of eight: the phone call.

Where home-service businesses lose jobs

Jobs leak in the gap between a customer reaching out and someone getting back to them, usually while the owner is on a roof or in a crawlspace. Measure your own leaks for two weeks before buying anything: the numbers tell you what to fix first and give you a baseline to judge results against.

Then put a dollar figure on the biggest leak, which for most shops is missed calls. Change the numbers to match your own:

That’s revenue before costs. To subtract the plan and your own time, use the calculator in our guide to AI receptionist costs in Canada.

Check the website too: a tap-to-call button at the top of every mobile page, and a quote form that lands in your CRM instead of an inbox nobody reads (see web design in Toronto).

One job, start to finish: how Lane works

Lane is the AI dispatch system we built for a Toronto home-services company with several service lines across the GTA. Here is one after-hours call traced from ring to review request. Each step is tagged by what does the work: AI (a language model deciding what to say), Automation (fixed code that does the same thing every time) or Person.

1. Automation

   9:42 PM: the phone rings. The business number is connected through the phone provider’s API, so an after-hours call starts a flow in Lane instead of going to voicemail.

2. AI

   The voice agent picks up. An ElevenLabs voice agent works out the problem, takes the details and books a window: tomorrow, 8 to 10 AM. If the caller needs a person, it hands the call over.

3. Automation

   The job writes itself. Customer, address, issue, priority and the call summary land in Lane’s CRM (Supabase). Repeat customers are recognized instead of becoming duplicates.

4. Automation

   The owner gets one Telegram alert. One message per new job, with a link into Lane. Contact details stay behind sign-in, not in the chat. A failed alert shows as failed and can be retried.

5. Person

   Next morning: dispatch. The job is assigned to a tech on the dispatch board.

6. Automation

   The customer gets an ETA. Nobody in the office has to phone them.

7. Person

   On site: WhatsApp closeout. The tech sends “Done” with a note and a photo from WhatsApp.

8. Automation

   The job closes and the review request goes out. The closeout triggers it, so nobody has to remember.

Count the AI tags: one out of eight. Everything after the call is ordinary automation, which is what you want for anything that sends a truck or messages a customer. If the agent mishears something, the damage stops at one booking that a person sees in the morning. What happens inside that one AI step is traced in our guide to what an AI agent is.

Our AI receptionist service covers the phone side of Lane. The CRM, alerts and closeouts are business process automation.

Can AI dispatch my techs?

Partly. Picking the tech is a rules problem, and rules are safe to automate. Judgment calls about the day stay with a person.

That’s how Lane runs. A model choosing which truck goes where adds nothing a written rule can’t do, and when it gets one wrong, you can’t see why.

Emergency calls: fixed rules decide

A language model is good at recognizing “there’s water coming through the kitchen ceiling” in a hundred phrasings. It can flag an emergency, but it should never be the thing that rules one out. When a call matches a trigger, the agent reads an approved safety line word for word and starts the emergency path. Here’s a starting rule set for a GTA shop. Adapt the wording with your team and your insurer.

Build in three rules:

1. Triggers run before booking logic. Match on keywords and on the AI’s own classification, and let either one start the emergency path. The AI can add an emergency. It can never downgrade one.
2. The safety line is fixed text. The agent doesn’t paraphrase it, and you approve every word.
3. Transfers have a fallback. On-call tech, then backup, then owner, with a text alert carrying the callback number if nobody picks up. The caller hears exactly what happens next.

Set the triggers wide. Waking the on-call tech for a dripping tap costs one phone call. Booking a gas smell for 8 AM is the mistake you can’t take back. The built-in tools escalate too: Jobber on terms you set (such as “emergency”), ServiceTitan by keyword, and Housecall Pro says its emergency handling is always on. Either way, write the triggers in your customers’ words: “smells like rotten eggs”, “water everywhere”, “sparks from the panel”.

Jobber, ServiceTitan or Housecall Pro: is the built-in AI receptionist enough?

If you run on Jobber, ServiceTitan or Housecall Pro, check your account before buying anything else. Vendors call it an AI receptionist, an AI CSR or a voice agent. It’s the same job. A built-in agent books into the schedule your techs already work from, with nothing to integrate or maintain. Here’s what each vendor’s documentation says as of October 2026:

None of the three help articles mentions French. If you take calls in French, ask for a test line before you sign.

Custom costs more up front and someone has to maintain it. A middle path: keep your field-service software as the system of record and build only the missing pieces around it, if your plan includes API access. That access decides whether a custom agent sees live availability or writes to its own CRM.

Six follow-ups to automate after the call (and what CASL allows)

Once the job is booked, six messages decide whether it gets done, paid and reviewed. All six run on ordinary automation (n8n, Zapier, Make or your field-service software’s own automations), with AI at most drafting the wording.

1. Quote follow-up that stops on reply. A nudge two days after the quote and one more later in the week, which matters most on furnace replacements and panel upgrades. Any reply stops the sequence and goes to a person.
2. Reminders and on-my-way texts. The arrival window goes out the day before. The tech’s name goes out when they head over.
3. Job closeout with photo. The tech’s note and photo close the job and give the customer a record: the new water heater’s rating plate, the repaired joint, the labelled panel.
4. Invoice and payment reminders. The invoice goes out at closeout, with reminders at set intervals and overdue accounts flagged to the owner.
5. Review request. Send it after every completed job, not only the happy ones. Google’s Maps content policy bans selectively asking for positive reviews and offering discounts in exchange for any review.
6. Seasonal maintenance-plan renewals. Furnace tune-ups before heating season, AC checks in spring, yearly water heater flushes on plumbing plans.

Each message falls somewhere different under Canada’s Anti-Spam Legislation (CASL), which covers email, texts and messaging apps:

Consent-exempt messages still need three things: your business name, a way to reach you and a working unsubscribe (s. 6(2)). Unsubscribes must be honoured within 10 business days (s. 11(3)).

Phone calls follow different rules

Sales calls fall under the CRTC’s telemarketing rules: check the National Do Not Call List and call only from 9 AM to 9:30 PM on weekdays and 10 AM to 6 PM on weekends, in the customer’s local time. An AI agent that dials out likely counts as an automatic dialing-announcing device, defined as equipment that can “convey a pre-recorded or synthesized voice message,” and sales calls through one need express consent. In 2019 the CRTC fined an Ontario HVAC company $194,330 for more than 96,000 calls to numbers on the list. Its June 2026 review asks whether these rules capture AI-generated voices, so check for a decision before running outbound campaigns.

General information, not legal advice. Have a lawyer review your message templates before launch.

SMS, WhatsApp or Telegram?

Pick by who’s on the other end. Customers need a channel that works without installing anything. Your techs need one that handles photos and links.

Lane splits it this way: Telegram for the owner’s job alerts, WhatsApp for technicians’ closeouts. For customers, start with SMS and add WhatsApp if they ask for it.

Privacy on the job

Under PIPEDA, which covers private-sector businesses in Ontario, personal information is anything about an identifiable person. For a plumber or electrician that includes addresses, door codes, call recordings, photos inside someone’s home and notes like “elderly, lives alone.” Alberta, BC and Quebec have their own similar laws.

If you record or transcribe calls, the federal privacy commissioner’s guidance on recording customer calls (page last modified October 8, 2026) expects you to say so at the start, name every purpose the recording may serve, give callers a choice and offer an alternative when recording isn’t essential. One greeting covers it:

Hi, you’ve reached the virtual assistant for [your company]. I can book a visit or get someone on the line. This call is recorded so we can book your job and keep accurate notes. If you’d rather not be recorded, say “callback” and a person will call you back.

List every purpose: if your voice vendor keeps recordings for quality checks or model training, the greeting has to say so. Opening with “virtual assistant” matters too. ServiceTitan’s own setup guide notes that telling callers they’re talking to an AI is a legal requirement in some places.

Then decide who sees what:

Owner and dispatcher

The full job record, behind a sign-in.

Techs

Address, issue and access notes for their own jobs, not the whole customer list.

Alerts

No phone numbers, addresses or door codes in Telegram or WhatsApp messages. Lane’s alerts carry a link, and the details stay behind sign-in.

Photos

Photograph the work, not the room.

Retention and vendors

Set how long recordings are kept, and check where each vendor processes data. Our AI receptionist vs answering service guide covers cross-border processing.

How to roll out an AI receptionist in a 2 to 15 tech shop

Start where the AI can only improve things: calls that would have gone to voicemail.

1. Write the rules. Your top call reasons, emergency triggers, booking windows, service area by postal code, and what the agent may say about prices.
2. Run 20 test calls before any customer hears it. Include each emergency trigger, a caller who wants a price, a caller who wants a human, a basement unit with a side entrance, a noisy truck cab, a French speaker and an existing customer asking where the tech is.
3. After hours only. Read every transcript for the first two weeks and fix what breaks.
4. Overflow. Daytime calls ring your team first. The agent answers after a set number of rings or when every line is busy.
5. First answer. Only once booking accuracy holds, with staff taking the transfers.

Review five numbers every week:

FAQ