How a Custom AI Voice Agent Books HVAC Dispatch 24/7
Most HVAC shops lose 15-25% of inbound calls. Not because nobody answers — the office manager is doing her best — but because the phone rings during a tune-up, the message gets missed, the customer finds someone else. By the time anyone listens to voicemail, the job is gone.
AI voice agents have crossed the threshold where they actually solve this. They answer in two rings, talk like a person, book the appointment, and dump the details into your CRM. We've been building them for home service shops for the last year — here's what works, what doesn't, and what it actually takes to deploy one.
What a working dispatch AI actually does
The version we've been deploying handles five categories of inbound call automatically:
- New service requests — collects address, equipment, symptoms, urgency. Books a same-week slot from the live schedule.
- Maintenance scheduling — looks up the existing contract, finds the next available slot in the customer's preferred window.
- Emergency triage — detects "no heat," "no AC," "gas smell," and routes to a real on-call tech immediately instead of trying to handle it.
- Status inquiries — "where's my tech?" questions get answered from the dispatch system without escalating.
- Quote callbacks — leaves a structured handoff for your estimator with everything needed to call back prepared.
What it sounds like
Modern voice models (we use the ElevenLabs and OpenAI realtime stacks) sound human. Customers regularly thank the AI by name at the end of the call and have no idea they were talking to software. The speed is the giveaway — it answers in under two seconds and never sounds rushed.
The one place we tell shops to be cautious: do not lie about it. If a caller asks "am I talking to a real person?", the agent should be honest. The trust hit from being caught lying is much worse than the novelty of being told upfront.
What it costs to run
Voice AI runs on a per-minute pricing model. Typical breakdown:
- Telephony (Twilio): ~$0.014/minute
- Speech-to-text (Deepgram / Whisper): ~$0.005/min
- LLM (GPT-4o / Claude): ~$0.02-0.04/min
- Text-to-speech (ElevenLabs): ~$0.05/min
Call it $0.10/minute all-in. A 5-minute booking call costs $0.50. If you take 30 calls a day, that's ~$15/day in raw API cost. The custom build itself runs $8,000-15,000 depending on integrations.
The tradeoffs nobody talks about
Two things to know before you commit:
You need a clean CRM first. The AI has to read from and write to your job records. If your data lives across spreadsheets and group texts, the AI can't dispatch from it. We almost always recommend a custom CRM build as the foundation before bolting on voice.
Some calls still need a person. Complex re-routing, unusual financing questions, irate customers — AI is great at the common cases and ordinary at the edge cases. The shops that win treat AI as the front line that handles 70-80% cleanly, and route the rest to a human.
The best voice deployments aren't trying to replace the office. They're letting the office stop being a switchboard so they can actually do the work that grows the business.
What it looks like deployed
For a 6-truck HVAC shop in the Midwest, we deployed a voice agent that:
- Answers every after-hours call within 2 seconds
- Books straightforward service in under 90 seconds
- Triages no-heat emergencies to on-call within 30 seconds
- Cuts missed-call lost revenue by an estimated 40%
- Saves the office manager ~20 hours/week of phone time
If you're running an HVAC shop and your office is drowning in phone traffic, this is the highest-leverage AI deployment available right now. See our Voice AI Agents page for the full build, or read more about custom CRMs for HVAC contractors that make voice deployment possible.