Key Takeaways
- An AI receptionist should ask safety questions first (gas smell, carbon monoxide alarm, smoke, water), then ask about who is in the home and the indoor temperature, then ask about the equipment.
- A true emergency is a safety hazard, a vulnerable occupant in extreme heat or cold, or a commercial system whose failure destroys stock or shuts the business. Everything else is urgent or routine.
- Write dispatch rules as plain if/then statements: safety hazard means instruct the caller and page the on-call tech now; vulnerable occupant means page now; no heat or cooling with no risk means book the first slot.
- The AI should never diagnose or promise an arrival time it cannot keep. It captures facts, gives scripted safety instructions, and hands off a clean summary.
- Test the rules with real past calls before going live, and review transcripts weekly for the first month.
What is an AI receptionist for HVAC emergencies?
An AI receptionist is software that answers inbound calls and texts in natural conversation, asks structured questions, books jobs, and alerts staff. For an HVAC company, the job is to sort a "no heat at 2 a.m." call into emergency, urgent, or routine, then trigger the right action without a person picking up the phone.
The value is in the sorting. A voicemail box tells you someone called. A triage script tells you whether the caller smells gas, has an infant in a 50 degree house, or simply wants a tune-up quote. InstantReply.ai handles this kind of call flow by running your triage questions, then paging the on-call tech or booking the visit based on the answers.
What counts as a true HVAC emergency?
A true HVAC emergency is a situation where waiting until morning creates a safety risk, a health risk, or a large property loss. Most no heat and no cooling calls are uncomfortable but not emergencies, which is why the AI needs defined tiers rather than treating every call as urgent.
| Tier | Examples | Default action |
|---|---|---|
| Emergency (safety) | Gas smell, CO alarm sounding, smoke or burning smell from equipment, sparking, water pouring near electrical | Read safety script, page on-call tech immediately, log the call |
| Emergency (vulnerable or loss) | No heat in freezing weather or no cooling in extreme heat with an infant, elderly, or medically fragile occupant; walk-in cooler or server room down | Page on-call tech immediately |
| Urgent | No heat or cooling, healthy occupants, tolerable indoor temperature | Offer first available slot, send text confirmation |
| Routine | Weak airflow, noises, thermostat questions, maintenance, quotes | Book standard appointment |
Your own thresholds will differ by climate and contract. A commercial customer on a service agreement may be an emergency at any hour, while a residential no-cooling call in mild weather may not be.
What should the AI ask first on a no heat or no cooling call?
The AI should ask safety questions before anything else: "Do you smell gas or burning?", "Is a carbon monoxide alarm going off?", and "Is anyone feeling dizzy, nauseous, or unwell?" These take seconds, and a yes to any of them changes the entire call.
If the caller reports a gas smell or an active CO alarm, the scripted response is simple. Tell the caller to leave the building, call 911 or the gas utility from outside, and not to switch anything on or off. The CPSC's carbon monoxide fact sheet gives the standard advice to get to fresh air and call 911, and the CDC lists the symptoms. Have your owner or safety lead approve the exact wording, and keep the AI to that wording rather than letting it improvise.
At the same moment, the AI should page the on-call tech with a short summary so the company knows a safety call is in progress.
Which triage questions separate a real emergency from a routine heating call?
For no heat, the questions that matter are the outdoor and indoor temperature, who lives in the home, and what the system is doing. Five or six questions are enough. More than that and callers hang up or get frustrated.
| Question | Why it matters |
|---|---|
| "What is the temperature inside right now?" | Separates uncomfortable from dangerous |
| "Is anyone in the home an infant, elderly, pregnant, or medically fragile?" | Triggers vulnerable-occupant escalation |
| "Is it a furnace, heat pump, boiler, or something else?" | Tells dispatch what parts and skills to send |
| "Is the system making noise, blowing cold air, or completely silent?" | Gives the tech a starting point |
| "Do you have another working heat source?" | Lowers urgency if the answer is yes |
| "Is this a home or a business?" | Commercial calls often follow different priority rules |
The AI should record answers as structured fields, not a free-text blob, so the on-call tech can read the summary in ten seconds on a phone.
What should the AI ask on a no cooling call?
For no cooling, the deciding factors are indoor temperature, the occupants' health, and whether the system protects something that spoils. A warm house in mild weather is routine. A 95 degree house with an elderly resident, or a failed walk-in cooler at a restaurant, is not.
Useful questions include:
- "What is the indoor temperature, and what is the forecast today?"
- "Is anyone in the building elderly, very young, or on medication that makes heat dangerous?"
- "Is the unit running but not cooling, or not running at all?"
- "Is this a home, an office, or a space with refrigeration or servers?"
- "Is there water leaking from the unit or ice on the lines?"
Water near electrical equipment moves a call up a tier. Ice on lines usually means the caller can switch the system off and wait, which the AI can suggest only if you have approved that instruction in advance.
When should the AI dispatch or alert the on-call tech?
The AI should alert the on-call tech immediately for any safety hazard, any vulnerable occupant with a failed system in extreme weather, and any commercial customer whose failure risks stock or operations. For everything else, it should book the earliest slot and confirm by text.
| Condition | AI action | Who gets contacted |
|---|---|---|
| Gas, CO, smoke, or electrical hazard | Read safety script, then escalate | On-call tech and owner, by call and text |
| Vulnerable occupant plus no heat or cooling in extreme weather | Escalate now | On-call tech |
| Commercial refrigeration or server room down | Escalate now | On-call tech |
| No heat or cooling, healthy occupants, tolerable temperature | Book first slot, state the after-hours fee if you charge one | Dispatcher in the morning |
| Maintenance, noise, quote request | Book standard visit | Office queue |
If the on-call tech does not acknowledge within a set window, the system should try a second person. Set that window and the backup order in advance, and make sure the caller is told the truth: "A technician has been alerted and will call you back" is better than a promised arrival time.
What details must the AI capture before handing off?
The AI must capture the caller's name, callback number, service address, equipment type, symptoms, and the triage answers. Missing any one of the first three turns a good lead into a callback guessing game.
Confirm the address by reading it back, and confirm the number by texting a message to it during the call. If the caller says they are at a different address than the one on file (a rental, a parent's home), record both. The summary sent to the tech should lead with the tier, then the key risk, then the address.
How should the AI handle callers who say everything is an emergency?
The AI should take the caller's urgency seriously, ask the same triage questions, and apply the same rules to every caller. If the answers do not meet the emergency threshold, it explains the next available slot and any after-hours charge, and offers to keep the caller on the early cancellation list.
Callers who disagree with the tier should have a path to a human. A simple rule works: if a caller asks twice for a person, page the on-call tech or office lead with a note that the caller wants to speak to someone. Your AI should not argue.
How does AI answering compare with a human answering service?
An AI receptionist such as InstantReply.ai answers every call instantly at any hour and applies your rules the same way each time, while a human answering service can use judgment on odd situations. Many HVAC companies use the AI as the first layer and keep a human for escalations.
Human services such as Ruby and Smith.ai are established options with live agents. The tradeoffs are cost per minute, hold times during peak weather events, and consistency of script use. During a heat wave or cold snap, call volume spikes, and an AI does not queue callers behind each other. Compare pricing and test both with your own scripts before you commit.
What does it cost when an emergency call goes unanswered?
An unanswered emergency call usually costs you the job, because the caller phones the next company on the list. The exact loss depends on your average ticket, but you can estimate it with your own numbers.
For example, say your after-hours emergency jobs average $600 and you miss four a month because the phone rings out. That is $2,400 a month, or $28,800 a year, in jobs that went to a competitor. Swap in your own ticket size and missed call count. The figure is an illustration, not an industry average.
How do you set up and test the triage rules?
Start by writing your tiers on one page, then convert each into an if/then rule the AI can follow. Run twenty to thirty past calls through the script and check that each lands in the tier a dispatcher would have chosen.
- List your emergency conditions and get the owner to sign off on the safety wording.
- Set the on-call schedule, the acknowledgment window, and the backup contact.
- Decide your after-hours fee wording and when the AI states it.
- Test with real past call recordings and a few staged calls.
- Review transcripts daily in week one and weekly for the first month, then tighten any question that confused callers.
With InstantReply.ai, routine calls then flow into the same system: booked appointments, a confirmation text, appointment reminders, and a missed call text back if a caller drops before finishing. After the job, the same platform can send review and referral requests and, later, maintenance reminders to reactivate cold leads.