AI

AI Phone Receptionists for Burnaby Trades This Winter

October 1, 2026 · 7 min read · Autana Solutions, Vancouver
AI Phone Receptionists for Burnaby Trades This Winter — Autana Solutions

It's 2:40 a.m. in January. A pipe has split in a crawlspace in South Burnaby, water is coming through the ceiling, and the homeowner is working down a list of plumbers on their phone. The first two go to voicemail. The third picks up. Guess who gets the job, and probably the repiping quote in spring too.

That's the whole case for an AI phone receptionist in the trades. Not that it's clever, but that it answers. This post walks through how these systems actually work, what the official documentation says they can and can't do, the privacy rules a BC business has to follow, and where we'd tell you not to bother.

What an AI receptionist actually does on a call

Strip away the marketing and a voice AI system is three parts stitched together: speech to text, a language model that decides what to say, and text to speech. Twilio's reference docs for its Conversation Relay product describe exactly this split. The service "handles the complexities of live, synchronous voice calls, such as speech-to-text (STT) and text-to-speech (TTS) conversions," and your application receives the caller's transcribed words and sends text back to be spoken (Twilio, ConversationRelay TwiML reference).

As of October 2026, that page lists Google and Deepgram as the speech to text options and Google, Amazon and ElevenLabs as the voice options. That's a moving target, so check the current page before you build on it. Two details matter for trades:

Jargon. A `hints` attribute lets you feed in "brand names, industry-specific terms, or other expressions" so the transcriber is more likely to catch them. Think "backflow preventer," "combi boiler," "Navien," "heat pump defrost." Without this, the AI will mishear the very words that tell you how urgent a call is.

Handoff. When the session ends, the app can pass back a reason. Twilio's own example payload reads "The caller requested to talk to a real person." That's the hook you use to transfer a live emergency to whoever is on call.

A sensible after-hours call flow

Here's the shape we'd recommend for a plumbing, heating or electrical shop in Burnaby or New Westminster. It's illustrative, not a template you should copy blind.

  1. Answer and disclose. "You've reached Smith Plumbing's after-hours line. I'm an automated assistant and this call is recorded so we can dispatch the right tech." More on why that second sentence matters below.
  2. Screen for life safety first. If the caller mentions a gas smell, rotten eggs, hissing near an appliance, sparking or smoke, the AI stops taking a booking and reads out the right instruction. For gas, FortisBC is unambiguous: stop what you're doing, don't use your cellphone or flip switches inside, go outside, and once outside call the FortisBC emergency line at 1-800-663-9911 (24 hours) or 911 (FortisBC, Gas leaks and odour). Your receptionist should never try to troubleshoot that call.
  3. Triage the real emergencies. No heat with kids or seniors in the house, active water leaks, a frozen line that's about to go. Collect the address, callback number, what's happening, and whether they can shut off the water or breaker.
  4. Page the on-call tech. Text and call the person on rotation with a short summary. If they don't acknowledge in a set number of minutes, escalate to the owner.
  5. Book the non-urgent stuff. A dripping tap at midnight doesn't need a truck. Offer the first morning slot and send a text confirmation.

The value isn't in step 5. Any booking widget does that. It's in steps 2 through 4, where a human dispatcher used to be the only option.

The privacy rules a BC trade has to follow

Most local trades aren't federally regulated, so the law that applies is usually BC's Personal Information Protection Act, not PIPEDA. The provincial guide is plain about it: PIPA covers "all private sector organizations in B.C.," including a person, a corporation and a partnership, and it requires organizations to "obtain your consent when they collect, use or disclose your personal information" and to destroy or anonymize it once it's no longer needed (Government of BC, Guide to the Personal Information Protection Act).

An AI receptionist collects a lot of personal information. Names, home addresses, phone numbers, sometimes the fact that an elderly parent lives alone with no heat. The federal privacy commissioner's guidance on recording customer calls is written for PIPEDA, but the page itself says that for businesses not covered by PIPEDA "the guidelines below are best practices" (Office of the Privacy Commissioner of Canada, Recording of Customer Telephone Calls). The key points:

  • Tell the caller the call is recorded and say why.
  • If they keep going knowing that, consent is implied.
  • Don't say "for quality assurance" if you'll actually use the recording for marketing.
  • Recordings can capture "incidental personal information offered by the caller not relevant to the call," so keep only what you need and limit how long you keep it.

Contractors carry this too. The OPC notes that businesses contracting out call centre services "must ensure that these third parties also follow the rules." An AI vendor answering your phones is in that seat, so ask where transcripts are stored and how long they're kept.

Who's responsible when the AI gets it wrong

You are. That isn't us being dramatic. Twilio's "AI Nutrition Facts" labels for Conversation Relay state under input/output consistency that the "Customer is responsible for human review." Using the feature also requires agreeing to Twilio's generative AI features addendum (Twilio, ConversationRelay TwiML reference).

For a structured way to think about that responsibility, NIST's AI Risk Management Framework is a solid public reference. It's voluntary, it was released January 26, 2023, and NIST followed it on July 26, 2024 with a Generative AI Profile aimed at the risks specific to systems like these (NIST, AI Risk Management Framework). NIST's page also notes, as of August 2026, that AI RMF 1.0 is being revised. You don't need to read all of it. For a phone receptionist, the practical takeaways are simple: write down what the AI is allowed to say, test it against nasty edge cases before it goes live, and review a sample of real calls every week.

Where this doesn't apply

We'd rather lose a sale than install something that hurts you. Skip it, or wait, if any of these sound like you:

You don't actually do after-hours work. If you won't roll a truck at 3 a.m., an AI that answers at 3 a.m. just collects messages a voicemail could take. A good voicemail greeting plus a text back in the morning may be enough.

Nobody is on call. The AI can capture the emergency. It can't fix the pipe. If the page goes to a phone that's on silent, you've just made a promise you can't keep, which is worse than not answering.

Your call volume is tiny. A shop getting two or three after-hours calls a month may be better served by a human answering service, or by the owner keeping the phone on the nightstand in cold snaps.

The technology still fails. Twilio's own docs include an example session that ended with "Network connection to WebSocket server failed." Speech recognition struggles with bad cell signal, wind, a panicking caller, or heavy accents. Every setup needs a fallback that sends the caller to a human or voicemail when anything breaks, and you need to test that fallback.

Safety calls aren't the AI's job. Gas, fire, electrical hazards and medical emergencies go to 911 or the utility. The receptionist's only role there is to say so clearly and fast.

How to start without overbuilding

Pick one season and one problem. For most trades in Metro Vancouver, that's winter emergency calls between 6 p.m. and 7 a.m. Run the AI on overflow and after hours only, so your daytime office keeps doing what it does well. Listen to every call for the first two weeks. Tune the hints list, the triage questions and the escalation timing. Then decide, on your own numbers, whether it earned its keep.

Sources

  • Twilio. "TwiML Voice: ConversationRelay." Twilio Docs, page updated September 2026. https://www.twilio.com/docs/voice/twiml/connect/conversationrelay
  • FortisBC. "Gas leaks and odour." FortisBC, accessed October 2026. https://www.fortisbc.com/safety-outages/energy-safety/natural-gas-safety/gas-leaks-and-odour
  • Government of British Columbia, Office of the Chief Information Officer. "Guide to the Personal Information Protection Act." https://www2.gov.bc.ca/assets/gov/business/business-management/protecting-personal-information/pipa-guide.pdf
  • Office of the Privacy Commissioner of Canada. "Recording of Customer Telephone Calls." Modified 2018. https://www.priv.gc.ca/en/privacy-topics/surveillance/02_05_d_14/
  • National Institute of Standards and Technology. "AI Risk Management Framework." NIST, page updated August 2026. https://www.nist.gov/itl/ai-risk-management-framework

If you run a trade in Burnaby, New Westminster or anywhere in the Lower Mainland and you're tired of losing winter emergency calls to whoever picked up first, book a free call with Autana. We'll look at your call patterns honestly, and if an AI receptionist isn't the right fit yet, we'll tell you that too.

AI receptionisttradesBurnabyafter-hours callsprivacywinter emergencies

Want an AI employee for your business?

We install a 24/7 AI worker for businesses in Vancouver, Burnaby, and beyond. Book a free Discovery Call.

Book a call →

Keep reading