The HVAC inbound call problem
HVAC companies miss calls. A lot of them. The pattern is always the same: a technician is on a job, the office coordinator is handling another customer, and the phone rings. If no one picks up, the caller books with whoever answers next.
In HVAC, the cost of a missed call is high. Emergency AC repair in July is not a price-sensitive purchase. The customer wants someone who picks up.
AI voice agents are a direct solution to this problem. Not a partial solution — a complete one for inbound call handling, at a cost that makes sense for a small HVAC business.
What the agent actually does
A well-configured AI voice agent for HVAC handles:
- Inbound call qualification — Is this an emergency (no cooling/no heat) or a maintenance request? Urgency routing matters here.
- Appointment scheduling — Direct calendar integration. The agent books the slot and sends a confirmation text.
- Basic troubleshooting questions — "Is your filter changed?" "Is the circuit breaker tripped?" Filtering obvious DIY issues reduces dispatches.
- Callback requests — When a customer wants to speak with a human, the agent captures their information and queues the callback.
What it does not do: negotiate price, diagnose complex problems, or handle warranty disputes. Those go to a human.
The deployment setup
We use a Modal-hosted LLM (currently a 14B model) combined with a voice layer (Twilio or a similar provider). Calls route through a phone number the HVAC company controls. The agent runs 24/7; the business hours logic is handled in the prompt.
Cold start on the LLM is the biggest technical challenge. First call of the day after inactivity can hit 60-180 seconds of load time. We handle this with a warm-up ping on a schedule. For inbound calls, the phone rings while the model loads — acceptable for most situations, borderline for emergency calls.
What it costs
At current Modal pricing, running a 14B model for a mid-volume HVAC company (150-300 calls/month) runs approximately $40-80/month in inference costs. Twilio adds ~$0.01/minute per call. Total infrastructure cost is under $150/month for most small operators.
Compared to a part-time answering service ($500-1000/month) or a missed call rate of even 5% on a $2,000 average job, the math is straightforward.
What breaks
Three failure modes come up consistently:
- Accent and background noise — Callers on a job site with equipment running create transcription errors. The agent should ask for confirmation on addresses and dates explicitly.
- Complex scheduling — "Can you do next Tuesday afternoon but not before 2pm and not after 4?" Multi-constraint scheduling with natural language needs careful handling. Better to offer slots than to confirm open-ended requests.
- Angry callers — Someone whose AC has been out for three days and is calling for the second time is not in a mood to interact with an AI. Fast human escalation path is non-negotiable.
Verdict
AI voice agents work well for HVAC inbound when scoped correctly. The scope is: capture the call, qualify the urgency, book the appointment or queue the callback. Anything beyond that should go to a human.
The ROI case is strong for any HVAC company doing more than 50 inbound calls per month and currently running on manual phone handling.