AI for Auto Shops: Book More Jobs in Burnaby

Every auto repair shop has the same leak. The phone rings while you are under a car with your hands full, it rings again at 6:40pm after the counter closes, and it rings on Saturday when one person is juggling three customers. Each of those calls was probably a booking. Most of them never call back. They call the next shop on the map.

That is the honest case for AI for auto shops. Not "transform your business." Just answer the phone every time, take the details, and put the job on the calendar.

What the adoption numbers actually say

Statistics Canada's Analysis on artificial intelligence use by businesses in Canada, second quarter of 2026 found 19.2% of Canadian businesses used AI to produce goods or deliver services in the 12 months before the survey, up from 12.2% in Q2 2025 and 6.1% in Q2 2024. Among businesses using AI, 28.2% reported using virtual agents or chat bots.

The size gap matters for an independent shop. That same release reports 27.8% adoption among businesses with 100 or more employees against 19.9% for businesses with 1 to 4 employees, and larger firms were far more likely to be running virtual agents or chatbots at 50.5%. Translation: the tooling your dealership competitor already has is now cheap enough for a three bay shop in Burnaby, and most small shops have not switched it on yet.

Where the booking actually leaks

Before buying anything, spend a week writing down where calls die. In most shops it clusters in a few places:

• Calls that ring out while techs are in the bays and the service advisor is on another line
• After-hours and weekend calls, which are heavily tow-in and "can you look at it Monday" work
• Quote requests that need a year, make, model and symptom before anyone can answer
• No-shows and vehicles that never come back for the deferred work you already quoted
• Callers who need Punjabi, Cantonese or Mandarin, which in Metro Vancouver is a real share of the phone

An AI voice agent only helps with the first three directly. The last two are follow-up problems, and they are usually solved with text, not voice.

What it actually costs to run

Model pricing is the part everyone guesses at, so here are published numbers rather than a vendor estimate. As of August 2026, OpenAI's API pricing page lists gpt-realtime-2.1 at $32.00 per million audio input tokens, $0.40 per million cached audio input tokens, and $64.00 per million audio output tokens. The smaller gpt-realtime-2.1-mini is listed at $10.00 and $20.00 per million audio input and output tokens.

OpenAI's Realtime API cost guide states the tokenization rate directly: "1 token per 100 ms of audio" for user input and "1 token per 50ms of audio" for assistant output.

So an illustrative five minute booking call, three minutes of customer talking and two minutes of the agent talking, works out to roughly 1,800 input tokens and 2,400 output tokens, or about 21 cents of model cost on gpt-realtime-2.1. That figure is illustrative and it is a floor, not a bill. The same guide notes "The entire conversation is sent to the model for each Response," so later turns cost more, and prompt caching only partly offsets it. Telephony minutes, transcription, and the integration into your shop management system are all on top.

Still, the shape is clear. The model is not the expensive part anymore. The plumbing and the setup are.

The rules you have to follow in BC

This is where most shops get sloppy, and it is the part with actual penalties attached.

Outbound reminder calls placed by an automated system fall under the CRTC's Unsolicited Telecommunications Rules. Even for non-solicitation calls like an appointment reminder, the message must "begin with a clear message identifying the person on whose behalf the telecommunication is made and a brief description of the purpose," and must give a mailing or email address plus a local or toll free number where a real representative can be reached. Calling hours are restricted to 9:00 a.m. to 9:30 p.m. on weekdays and 10:00 a.m. to 6:00 p.m. on weekends. The key rules summary is blunter about sales: an automated dialing announcing device may only be used for a solicitation call where "the consumer has given their express consent prior to the telemarketing call." A reminder that your brakes are due is one thing. An automated call pitching a tire promotion is a different legal category.

Recording is the other trap. The Office of the Privacy Commissioner's guidance on Recording of Customer Telephone Calls says an organization must inform the customer that the call is being recorded and state the purpose, that "The organization can only record a call for purposes that a reasonable person would consider appropriate under the circumstances," and that customers who object should be offered an alternative such as visiting the shop in person or completing the transaction online. If your AI agent transcribes calls, and it does, you are recording. Say so at the top of the call.

Where this doesn't apply

Some shops should not do this yet, and it is worth being specific about which.

If your call volume is genuinely low, say under 15 or 20 calls a day, and your advisor already catches nearly all of them, an AI phone agent is solving a problem you do not have. Fix your Google Business Profile and your online booking link first. Cost was cited as a barrier by only 10.6% of businesses in the StatCan release, which suggests the real blocker for small firms is fit, not price.

If your shop management system has no usable API, the agent cannot see real bay capacity. It will book a diagnostic into a slot that is already full, and your advisor now has two jobs instead of one. Check the integration before you check the model.

Diagnostic conversations are also a poor fit. A caller describing a grinding noise on right turns needs a technician's judgment, not a booking script. The safe design routes those to a human quickly rather than pretending to triage them.

And it needs supervision. The NIST AI Risk Management Framework 1.0 is explicit that mechanisms should be "in place and applied" to "supersede, disengage, or deactivate AI systems" that show inconsistent performance, and that system behaviour "is monitored when in production." In practice that means someone at the shop listens to five calls a week and has a switch that sends everything back to voicemail. If nobody owns that, do not deploy it.

A sensible first step

Start with one narrow job: answer calls that ring more than four times or come in after close, collect name, phone, year, make, model and the reason for the visit, and either book into a real open slot or text the advisor a callback task. Run it for 30 days and compare booked jobs against the same month last year. If the number does not move, you learned something cheap.

If you run a shop in Burnaby, New Westminster or anywhere in Metro Vancouver and you want a straight answer on whether this fits your setup, Autana Solutions will look at your call volume and your shop management system and tell you honestly if it is worth it. Book a free call with us and we will map it out with you, no pitch required.

Sources

• Statistics Canada. *Analysis on artificial intelligence use by businesses in Canada, second quarter of 2026*. 2026. https://www150.statcan.gc.ca/n1/pub/11-621-m/11-621-m2026010-eng.htm
• OpenAI. *API Pricing*. Accessed August 2026. https://developers.openai.com/api/docs/pricing
• OpenAI. *Realtime API: Managing costs*. Accessed August 2026. https://developers.openai.com/api/docs/guides/realtime-costs
• Canadian Radio-television and Telecommunications Commission. *Unsolicited Telecommunications Rules*. https://www.crtc.gc.ca/eng/trules-reglest.htm and *Key Unsolicited Telecommunications Rules*. https://crtc.gc.ca/eng/phone/telemarketing/tobligations/rules-regles.htm
• Office of the Privacy Commissioner of Canada. *Recording of Customer Telephone Calls*. https://www.priv.gc.ca/en/privacy-topics/surveillance/02_05_d_14/
• National Institute of Standards and Technology. *AI Risk Management Framework (AI RMF 1.0)*. 2023. https://airc.nist.gov/airmf-resources/airmf/5-sec-core/