AI Receptionist Setup in 14 Days: HVAC Guide
Week-by-week setup for HVAC contractors. Covers 8 common call types, exact scripts, first-week problems, and how to catch them early.
Kevin Sorrell runs a 7-truck HVAC shop in Frisco, TX. He signed up for an AI receptionist platform in early April 2025 and went live on April 22nd, two weeks before the first heat advisory of the season. By May 15th, his system was handling 62% of inbound calls without any human involvement. His office manager handled escalations, appointment changes, and warranty calls.
The reason it worked was the setup. Kevin treated the 14-day window seriously and caught two configuration problems before the platform touched a live customer call.
This is the playbook he used.
Operator details anonymized. Based on a real LeadExploder account matching this profile.
Day 1 through 3: the call script audit
Before you configure anything, you need to know what you are configuring. This step is skipped constantly, and it is why so many AI receptionists get launched with a generic script that does not match how the business actually operates.
Pull your call logs from the last 90 days. Listen to 30 recorded calls if your system has them. If not, have whoever answers your phones write down every call type they handled this week.
For HVAC, the eight most common inbound call types are:
- Emergency AC or heat failure (system completely out)
- Emergency calls where the unit is running but not performing
- Routine maintenance scheduling (biannual tune-up)
- New system install inquiry (homeowner is replacing equipment)
- Warranty or recall callback
- Existing appointment reschedule or cancellation
- Billing or invoice question
- General inquiry or referral (someone was referred and wants to learn more)
Types 1 and 3 will make up 60% to 70% of your total volume. Types 4, 5, and 6 are each around 8% to 12%. Types 7 and 8 are a small slice but need a path so the AI does not dead-end on them.
For each call type, document: what does the caller need to tell us, what do we need to book or dispatch, and what is the next action after the call.
Day 3 deliverable: a one-page document with all eight call types, the qualifying questions for each, and the outcome action for each. This is your configuration brief.
For a complete reference on how AI receptionist for HVAC handles the full range of call types in practice, that post covers each call path in detail.
Day 4 through 7: AI configuration and test calls
With the configuration brief ready, the platform build is straightforward. Here is the call script structure for the four highest-volume call types.
Emergency AC or heat failure:
“Thanks for calling [Company Name], this is [AI name]. Are you calling about an AC or heating emergency, or would you like to schedule service?”
[Emergency]: “Got it. Is the system completely off, or is it running but not cooling the way it should?”
[Completely off]: “And is there anyone in the home right now who is elderly, very young, or has a medical condition we should know about? [Pause.] I’m going to get this to our dispatch team right now. What’s the service address? I’ll send you a text with an ETA as soon as we confirm a tech.”
[Running but struggling]: “How long has it been like this? [Pause.] Okay, let me check availability. We can typically have someone out [today / tomorrow] — what’s the address?”
Routine maintenance scheduling:
“Of course. Are you already on a maintenance plan with us, or is this the first time scheduling?”
[Existing customer]: “Great. What unit are we looking at, and do you have a preferred timeframe?”
[New customer]: “No problem. What’s your zip code? I’ll check which tech covers your area and what times are open. [Pause.] We have [day] between [time range] and [day] in the afternoon. Which works better?”
New install inquiry:
“Happy to help with that. Is this replacing an existing system or a new installation? [Pause.] And is it residential or commercial? [Pause.] I’m going to get you scheduled for a free estimate with our install team. What’s the best address and a day that works for you this week or next?”
Warranty callback:
“Got it. Can you give me the unit model or the job number from your paperwork? [Pause.] And what’s the issue you’re experiencing? I’ll create a warranty ticket and have someone reach out to confirm the visit.”
By day 5, have your platform vendor walk you through all eight call flows in a test environment. On day 6, call your own business number from an external phone and run through each call type yourself. Take notes on every place where the response felt off or incomplete.
Day 7: fix the issues you found. Do not go live with unresolved issues. The most common fixes at this stage are the emergency qualifying question (making it clearer), the scheduling logic (connecting to live availability vs. placeholder times), and the SMS confirmation message (making it sound less automated).
Integration testing before go-live: how to run test calls through the system
Most HVAC operators skip this step and pay for it in the first week live. Integration testing means verifying that every data handoff works correctly before a real customer is on the other end.
Here is the integration test protocol Kevin ran on days 6 and 7:
Test 1: Emergency dispatch chain. Call your own business line from an external phone. Describe an emergency AC failure with a vulnerable person in the home. Confirm: Does the AI route to the emergency path? Does the dispatch alert fire to the on-call tech’s number within 60 seconds? Does the CRM record appear with the correct address, damage description, and urgency flag? Does the caller (you, in this case) receive an SMS confirmation within 2 minutes?
Test 2: Routine booking with calendar integration. Call and request a maintenance appointment. Confirm: Does the AI offer real available slots from your live calendar, not placeholder times? Does booking the slot block it in your dispatch software so no one can double-book it manually? Does the job record appear in ServiceTitan or Housecall Pro with the correct fields populated?
Test 3: The fall-through path. Say something the AI is not trained to handle: “I want to talk to someone about a billing dispute from three months ago.” Confirm: Does the AI route to the human escalation path? Does your office manager or dispatcher receive an alert? Does the call transfer correctly if transfer is configured?
Run each test twice. If the same test fails twice, the configuration has a gap. Do not go live with a known gap.
Day 8 through 14: live with monitoring
On day 8, the AI goes live. Every call is real.
This week is not set-and-forget. It is active monitoring. Here is what to watch.
Check the call log every morning for the first 5 days. Your platform should log every call with a transcript or summary. Read through the calls from the prior day. You are looking for three things: calls where the AI misclassified the call type (routed an emergency as a routine call), calls where the caller got frustrated and hung up early, and calls where the AI offered a time slot that did not match your actual schedule.
Have your office manager listen to 5 live calls on day 8 and day 9. Fresh ears catch things that you miss when you built the system. The goal is two rounds of quick fixes before the end of the second week.
Set a forwarding rule for calls the AI cannot resolve. Every AI receptionist should have an escalation path: if the caller says “I need to speak to a person” or if the call falls outside any trained scenario, it transfers to your office line or sends an immediate text to your manager. Confirm that this escalation path is working on day 8 with a test call.
The most common live-launch issue for HVAC setups is the emergency after-hours dispatch path. On a Monday morning, it works because someone is in the office to receive the dispatch alert. At 11 PM on a Saturday, the chain needs to work automatically. Test it specifically. Call your after-hours number after your office closes on day 8 and confirm that the on-call tech receives the alert within 60 seconds.
What to monitor in the first week live
Beyond the morning call log review, here are the metrics to track in week two:
Escalation rate: What percentage of calls are being handed off to a human? A well-configured system for HVAC should escalate 15% to 25% of calls. If it is above 40%, the AI is not handling enough call types. If it is below 10%, check whether the human escalation path is actually triggering when it should.
Emergency dispatch accuracy: For every emergency call in the log, confirm that the dispatch alert fired and the CRM record classified it as an emergency, not a maintenance booking.
Slot accuracy: Spot-check 10 AI-booked appointments against your dispatch calendar. Confirm the times match. If you find discrepancies, the calendar integration has a sync delay or a permission issue.
SMS delivery rate: Check your platform’s outbound SMS log. Every booked call should have a corresponding outbound confirmation text. If delivery is below 95%, there may be a carrier filtering issue on certain number types.
Also track where to find set up missed-call text-back instructions if your platform needs the text-back flow configured separately from the AI answering component.
What does a bad launch look like and how do you catch it?
The two most expensive launch failures for HVAC:
An emergency call gets classified as a routine booking. A homeowner with no AC in July calls in a panic and the AI books them for a maintenance slot on Thursday. They hang up and call a competitor who dispatches someone today. This happens when the qualifying question at the start of the call is too vague. “Is this about scheduling service or something else?” is too vague. “Is your system completely off right now?” is specific enough.
The dispatch integration creates a record with incomplete data. Your tech shows up to an address with no information about what the problem is or what equipment is in the home. The customer is irritated. This happens when the AI is not collecting enough detail in the call, or when the field mapping between the AI and your dispatch software is misconfigured. Check every field in the first three job records that the AI creates. If the “problem description” field is blank, the integration is not passing that data.
Both of these are catchable in the Day 6 test call review if you are thorough.
A real example from a Pearland, TX shop in summer 2024: The emergency escalation path was configured correctly for business hours, but the after-hours on-call tech SMS was set to send to the office manager’s phone instead of the rotating on-call number. On a Friday night at 10 PM, a homeowner with a failed AC unit called, the AI completed the intake correctly, and the dispatch alert fired to the office manager’s phone. The office manager was off the clock and did not see the message until Saturday morning. The job went to a competitor who had an on-call tech actually receiving alerts. That single configuration error cost a job.
The fix: test the after-hours dispatch path specifically, from an external phone, after your office closes, on day 8. Not during business hours. After hours. That is when the failure modes live.
Troubleshooting checklist
If something is not working in week two, work through this list before calling your vendor:
- Did the CRM integration credentials expire or get refreshed after setup? (Common when IT refreshes OAuth tokens)
- Is the AI offering time slots from the correct calendar? (Check whether the calendar connected is the right technician’s schedule, not a shared calendar with different permissions)
- Are emergency trigger phrases in the script specific enough? (“Something is wrong” is not a trigger. “The system is completely off” is.)
- Is the on-call dispatch SMS going to the right number and not a generic office number?
- Is the bilingual path configured if your market has Spanish-speaking callers?
- Is the human escalation path configured and tested, or just assumed to work?
What does Kevin’s outcome look like 30 days in?
Kevin’s numbers at the 30-day mark:
- Total inbound calls handled: 214
- Calls resolved fully by AI without human involvement: 133 (62%)
- Emergency calls dispatched: 28, with an average AI-to-dispatch time of 47 seconds
- No-show rate on AI-booked appointments: 9% (down from 17% before, because SMS confirmations went out within seconds of booking)
- Platform cost: $397 per month
The number Kevin cares about most: his office manager now handles approximately 40 calls per week instead of 90. She spends the other time on outbound follow-up, maintenance plan renewals, and the kind of customer relationship work that requires a human.
What to do this week
Start the call script audit. Block two hours, pull your call logs from the last 90 days, and list every call type your team handled. If you do not have call logs, call your own business line and see what happens. Then map each call type to the four fields: what the caller needs, what you need to collect, what the next action is, and whether it is an emergency or a scheduled service.
That document is the foundation of a setup that works. Everything else is configuration.
Book a demo and see the 14-day HVAC setup running live.
Alex Rocha is the founder of Mastodon Marketing, a Houston-based growth agency that runs marketing for service businesses across 70+ client sites. He built LeadExploder as the operating system he wished his clients had on day one. Learn more about Alex →
Frequently asked questions
How long does it actually take to set up an AI receptionist for an HVAC company?
A complete setup, including call script configuration, integration with your dispatch software, and live testing, typically takes 10 to 14 days when done carefully. Rushing the setup is the most common cause of a bad launch. The 14-day timeline includes a test period before the AI handles live calls.
What are the most common HVAC inbound call types an AI needs to handle?
The eight most common are: emergency AC or heat failure, routine maintenance scheduling, new system install inquiry, warranty or recall callback, existing appointment change or cancellation, billing or invoice question, filter replacement reminder response, and referral or general inquiry. Emergency calls and routine bookings together make up roughly 70% of total inbound volume.
What can go wrong when launching an AI receptionist for HVAC?
The three most common issues are: the AI misclassifying an emergency as a routine call (fixable with better qualifying questions), the dispatch integration failing to create a job record (usually a credentials or field-mapping issue), and the AI offering unavailable time slots because it is not connected to the live schedule. All three are catchable in test calls before go-live.
Do I need to rewrite my entire call script to use an AI receptionist?
No. Start with the scripts you already use for your team. The AI configuration process translates those into a structured call flow. The main addition is a qualifying decision point at the start of each call that routes the caller to the right script based on their need. That routing logic is what most HVAC shops are missing today.
