An AI phone agent can make sure every caller is answered in the first heat wave or cold snap, sorted by urgency and told an honest time window. It cannot add a technician. If your trucks are already full, the agent makes the queue orderly, not shorter.
That distinction decides whether it is worth buying for an HVAC shop. Below: what changes on the phones during a weather spike, what the agent should do, and when you should skip it.
What happens to an HVAC phone line in the first hot or cold spell?
More calls arrive at once, and more of them land outside office hours. Both break a small front desk.
ServiceTitan published off-hours call shares for residential HVAC shops on its platform. In its 2025 data, 14.1% of calls arrived outside business hours in June, against 9.8% in October. The biggest cluster fell between 5pm and 9pm.
Two limits. The data is not nationally representative, and it says nothing about cold snaps. I found no verified winter figure, so I give none.
Local news adds one data point. A technician quoted by KCTV5 in September 2026 said business was up at least 20 percent compared to a normal summer. That is one company in one city, not a trend line.
The practical result is the same everywhere. Your phone rings while the owner is on a roof, the dispatcher is on another line, and the third caller gets voicemail.
What should the agent do on a surge day?
Four jobs, in this order. This is what I'd require of any agent you hire, mine included.
- Answer every call, fast. My agent answers in about 1 second. In the Blue Ridge recovery demo, a missed-call text goes out in 8 seconds and the AI callback reaches the caller in 52.
- Triage by risk, not by who called first. Ask the questions that separate "AC is noisy" from "no cooling and an 80-year-old lives here".
- Book or promise a window. Book the slot if the calendar has one. If it doesn't, say so plainly and say when a person will call. Only promise a time your shop keeps.
- Leave a record of every call. The Blue Ridge demo is a single-prompt Retell agent on GPT-4.1 that must call a
log_outcomefunction before it hangs up, so every call has an outcome on file. That is a prompt rule, not a guarantee. The morning dispatcher can then start from those logged outcomes.
Notice what is missing: diagnosis. The agent doesn't tell a caller what is wrong with the unit, and it doesn't quote a repair price. My plumbing and HVAC page says it quotes the dispatch fee and flat-rate items it is given, and nothing else.
Which callers need a person right away?
Anyone at medical risk from the temperature. The CDC says people aged 65 or older are more prone to heat-related health problems, and its caretaker checklist asks whether the person has access to air conditioning. In cold weather the same logic applies to heat, though I have no official source to quote for it.
So the first triage question after "what's wrong?" should be "who is in the home?". Write the rule yourself, before the season:
- No cooling or no heat and an older adult at home, or anyone you know has a medical condition (my own judgment, not CDC guidance): transfer to the on-call person now, or book the first emergency slot.
- No cooling or no heat, healthy adults: book the next real slot and say when it is.
- Noise, a weak airflow, a thermostat question: book normal service.
- Gas smell, burning smell, carbon monoxide alarm: safety instruction first, then end the call. This is covered in how an AI phone agent handles emergency calls.
Test those lines. My test suite runs 38 simulated callers, 114 calls per run, for about $0.25 a run.
Its first run, on a cheaper model, missed 5 emergency handoffs in 41 plumbing calls. That was plumbing, not HVAC, and I haven't run an HVAC suite.
But the lesson carries over: an urgent handoff is the first thing to fail. Run the "elderly resident, no AC" call at least 10 times before the next spell. More on the method in how I test AI phone agents.
Does an AI agent help if you are already booked out?
Yes, in one narrow way. It stops you losing the caller to a worse experience than a wait.
Picture it: the calendar is full until Thursday. A caller with no AC hears voicemail and dials the next shop. With an agent set up as above, she hears "the first slot is Thursday morning, and a person will call you by noon", which she can accept or refuse, if that is a promise your shop keeps. If she refuses, you lose the job anyway, but you know it and you didn't make her wait on hold.
What it doesn't fix is capacity. Callers who need a repair today and can't get one will still go elsewhere. Don't buy an agent expecting it to create trucks.
What about the calls you place before the season?
That is where I'd start, honestly, because it is calm work. My plumbing and HVAC page describes a reactivation flow: it works your customer list and offers maintenance checks before the cold snap. That flow is wired to your customer list on a client build. It is not in the Blue Ridge demo, so you can't hear it there.
The spring and fall tune-up call is also a safer first project than the emergency line. It runs in daylight, nobody is in danger, and a mistake costs you an awkward conversation, not a missed emergency.
When is an AI phone agent not worth it for HVAC?
- You are a one-truck shop and answer your own phone. Voicemail plus a fast callback may be enough. The $900 build and $450 a month won't pay back at a few calls a week.
- Nobody owns the on-call phone. The transfer target has to answer. If it doesn't, the caller is worse off than with voicemail.
- Your dispatch rules aren't written down. The agent follows rules. If your team argues about who counts as urgent, settle that first.
- Most of your calls are commercial or contract work. Those customers have a named contact. Route them to that person.
- Your problem is capacity, not answering. If calls are answered and you simply can't service them, fix scheduling and staffing.
The pricing page lists 500 minutes a month, then 15 cents a minute. A surge month can use more than that. Ask what a month at twice normal volume would cost, and get the number in writing. Prices are on the pricing page.
What should you do before the next spell?
- Pull your phone log for last year's worst week. Count answered, missed and voicemail calls.
- Write the urgency rules: who counts as at risk, what gets transferred, what waits.
- Pick one on-call number, and ring it yourself at 9pm to see who answers.
- Decide what the agent says when the calendar is full, in one honest sentence.
- Run 10 test calls per urgency rule, and read the first 20 real calls by hand.
If you do only the first two steps, you will know whether an agent would have helped last summer. When you want to hear how a missed-call callback sounds, the recovery demo is live.
Originally published at adelatya.com.
