Do not buy another lead mill until your answer rate, first-touch time, and estimate follow-up are measured. Booking rate beats lead volume.
The HVAC leak is speed, not “more leads”
An HVAC lead is a person with a hot house, a dead furnace, or a replacement quote sitting on the kitchen counter. They called you, filled a Google Local Services form, or left a website request. They will call the next shop on the map if you are slow.
Most roundups fail this trade. They rank copy tools and quiz funnels next to answering tools. A writer does not recover a missed Saturday no-cool.
Invoca's 2022 Buyer Experience Report (cited in their home-services missed-call writeup) found 62% of buyers call before they purchase, 76% will stop after one bad experience, and over 60% will pay more for good service.3 HVAC is worse than that average on urgency. The buyer is not browsing. They are sweating.
A separate 2011 Harvard Business Review audit (Oldroyd, McElheran, Elkington) sent test web leads to 2,241 U.S. companies. Among firms that responded within 30 days, average response was 42 hours. 23% never responded. In a companion dataset of 1.25 million leads, firms that tried contact within an hour were nearly 7x as likely to qualify the lead as those that waited even one more hour, and more than 60x as likely as those that waited 24 hours.2
Keep the two studies apart. MIT 2007 is 100x contact / 21x qualify at 5 minutes vs 30 minutes on B2B web leads.1,19 HBR 2011 is the 42-hour average, the 23% never-respond, and the 7x within-an-hour figure.2 Neither number is an HVAC close rate. Use them as speed physics, then run your own phone log.
Simpro's 2025 Trades Outlook (600+ trade and field-service businesses) found 69% of owners see AI's biggest impact in workflow optimization, not in magically creating demand.5 That matches how HVAC shops should buy: plug leaks in the office, then spend more on ads.
A writer does not recover a Saturday no-cool.
The first AI that pays is the one that texts the missed ring and hands a human the thread.
What “AI lead generation” actually means in an HVAC shop
Four jobs. Rank them by cash recovered per week of setup.
- What the AI does
- Answers or texts back the inbound call/form in seconds, then hands a live person the transcript
- When it is first
- Always first if you miss calls or sit on LSAs
- When it is a waste
- If a dedicated CSR already answers in one ring and books
- What the AI does
- Runs a 5–12 touch SMS/email sequence on unsold quotes
- When it is first
- Second, once inbound is caught
- When it is a waste
- If you already close most changeouts in the house
- What the AI does
- Texts a Google review request at job complete
- When it is first
- Third, once jobs are actually run
- When it is a waste
- If you already have enough map-pack reviews to rank
- What the AI does
- Quizzes, chat widgets, D2D scoring, content
- When it is first
- Last
- When it is a waste
- If the office still dumps voicemail
| Job | What the AI does | When it is first | When it is a waste |
|---|---|---|---|
| First touch | Answers or texts back the inbound call/form in seconds, then hands a live person the transcript | Always first if you miss calls or sit on LSAs | If a dedicated CSR already answers in one ring and books |
| Estimate follow-up | Runs a 5–12 touch SMS/email sequence on unsold quotes | Second, once inbound is caught | If you already close most changeouts in the house |
| Review ask | Texts a Google review request at job complete | Third, once jobs are actually run | If you already have enough map-pack reviews to rank |
| New demand | Quizzes, chat widgets, D2D scoring, content | Last | If the office still dumps voicemail |
Ignore tools that only write service-area blogs. Eight honest HVAC pages you maintain beat fifty AI posts you will not stand behind on a callback. Start from the HVAC industry page if you need the trade-shaped version of this stack.
Protocol: buy in this order
Do this before you demo a sixth vendor.
Audit one week of phones
Export the call log from your carrier, Jobber / Housecall Pro / ServiceTitan, or call tracking. If you cannot produce this sheet, you are guessing. AI will not fix a shop that cannot count rings.
- timestamp
- answered / missed / voicemail
- duration
- after-hours flag (your posted hours)
- source if you have tracking numbers (Google LSA, GBP, website, truck wrap)
- Score: answer rate during posted hours, after hours, median minutes to first human contact, booked jobs from those contacts
First-touch SLA
Set a written rule the office can screenshot. Inbound call: live answer, or text-back in under 60 seconds, then a human in the thread. Google LSA / web form: first human or AI text in under 5 minutes (the MIT window, used as a speed target, not a close-rate promise).1 After hours: same text-back. Owner or on-call gets a push with address, system age, and heat/cool status. Do not let the bot quote a changeout.
Missed-call text
Send from the shop name, not a short code that looks like spam. The point of the AI is to fire that first text without waiting for a tech to climb off a roof. The point of the human is to book, set expectations, and not sound like a phone tree. See missed-call text-back for how the handoff is supposed to feel.
Hey, this is {Owner} at {Shop}. We just missed your call. Is this a no-cool / no-heat, or can it wait until morning? Reply here and we'll get you on the board.Still here. If you want a same-day window, send the address and the make/age of the system if you know it.
Estimate follow-up (day 2)
Hatch's HVAC sample was a campaign that starts two days after the estimate. The top campaign in that 163k set used 7 messages (5 texts, 2 emails) over 7 days with 2 pause days. First text ~200 characters. Average response across the set was 60%.12 Do not copy a 12-touch sequence if you will not have a salesperson in the thread. AI that pings and never hands off trains people to ignore you.
{First name}, it's {Rep} from {Shop}. Checking in on the {system} quote from Tuesday. Want me to hold a install window this week, or do you still have questions on the options?Review ask at complete
Same day the tech leaves, not Friday batch.
Thanks for having us out. If we earned it, a quick Google review helps other homeowners find a shop that actually shows up: {GBP short link}Only then buy more demand
Google Local Services, search ads, and a tight GBP beat Angi-style networks for a shop that already answers. If answer rate is still in the Invoca 27% neighborhood, more ads buy more misses.3
Tool cards, grouped by the leak they fix
These are not a ranked “top 10.” Fit the tool to the leak you measured.
Nobody picked up
Missed-call text-back
- Fit
- 1–8 trucks, the owner still lives on the phone, and you want the caller in a text thread before they dial the next pin on Maps.
- Caveat
- A text is not a book. Someone still has to qualify and put it on the board.
Jobber AI Receptionist
- Fit
- Shops already living in Jobber who want native booking. See Toolbench vs Jobber if the question is front office vs the job OS.
- Caveat
- It is Jobber's ecosystem. Escalation keywords (no heat, no cool, gas smell) have to be set and tested by you. A generic bot that books a “tune-up” on a 98° no-cool is worse than voicemail.
ServiceTitan AI Voice Agent
- Fit
- Shops already on ServiceTitan with enough call volume that overflow and after-hours are a desk problem.
- Caveat
- Wrong first buy for a two-truck shop that does not run Titan.
Podium AI Employee
- Fit
- Shops that already text customers and want reviews + inbox + AI CSR in one vendor.
- Caveat
- Price stacks. Audit the AI add-on separately from the review product.
Dedicated AI answering
- Fit
- After-hours coverage without ripping out dispatch.
- Caveat
- Whatever you pick, the handoff rule is the product: the owner or dispatcher sees the thread and owns the book. Voice AI as the only first touch on a high-ticket no-heat is how shops get one-star reviews that mention “I talked to a robot.”
The quote died in the folder
Hatch
- Fit
- Replacement-heavy HVAC with a real estimate volume and a salesperson who will jump into the thread.
- Caveat
- If you only run service tickets, this is later.
Map pack is thin, so inbound is thin
Podium, NiceJob, BirdEye
- Fit
- A shop that actually completes jobs and can ask once, from the shop number, with an opt-out.
- Caveat
- Pick one. Do not stack three review vendors.
You actually need more demand (last)
Google Local Services Ads + call tracking
- Fit
- A shop that already answers, with answer rate out of the Invoca neighborhood.
- Caveat
- More ads on a 27% miss rate buy more misses.
Door-to-door scoring
- Fit
- Replacement canvassing already on the calendar.
- Caveat
- Do not buy this to avoid picking up the phone.
Website chat / quiz funnels
- Fit
- A site that already gets form volume you can prove in the phone audit.
- Caveat
- A quiz is not a first-touch SLA.
Will AI replace HVAC technicians?
How to choose without a 40-tab spreadsheet
Ask every vendor these six. If they cannot demo them on a fake no-cool, walk.
- 01What happens in the first 60 seconds? Live voice, SMS, or both. Who owns the next message, AI or a named human?
- 02HVAC intake. System type, age, heat vs cool vs IAQ, address, service area, after-hours premium. If the script is generic “home services,” it will book dumb jobs.
- 03Emergency path. No heat, no cool, gas, carbon monoxide mention → ring the on-call cell, do not add to tomorrow's board.
- 04Where the record lands. Jobber, Housecall Pro, ServiceTitan, or a text to the owner. A lead that lives in a second CRM is a missed job with extra steps.
- 05After-hours behavior. Quiet hours for marketing texts. Always-on for true inbound. Mixing those is how you get TCPA problems and blocked numbers.
- 06Who tunes it after week one. If the answer is “you, in this portal, forever,” budget owner time. Native FSM tools (Jobber, ServiceTitan) win on less glue. Independent answering tools win on staying out of your dispatch.
Shop size heuristic
- Buy this
- Missed-call text-back + owner in the thread. Add reviews.
- Ignore this
- Enterprise voice until the log says you cannot keep the SLA.
- Buy this
- Overflow + after-hours voice, estimate follow-up, review automation.
- Ignore this
- Skipping the booking-rate comparison of bot vs human on the same job types.
- Buy this
- Native voice agent against Adaptive Capacity so you do not double-book.
- Ignore this
- A second AI receptionist that cannot see the board.
| Shop | Buy this | Ignore this |
|---|---|---|
| 1–5 trucks | Missed-call text-back + owner in the thread. Add reviews. | Enterprise voice until the log says you cannot keep the SLA. |
| 6–20 trucks with a CSR | Overflow + after-hours voice, estimate follow-up, review automation. | Skipping the booking-rate comparison of bot vs human on the same job types. |
| Titan-class | Native voice agent against Adaptive Capacity so you do not double-book. | A second AI receptionist that cannot see the board. |
Does AI lead generation work?
Yes, if you define the job as converting demand you already bought. No, if you define it as a robot that finds homeowners while you ignore the phone.
Evidence that is actually about response, not magic:
- MIT 2007: 5 minutes vs 30 minutes, contact 100x, qualify 21x, B2B web leads, 15k+ leads / 100k+ attempts / 6 companies.1,19
- HBR 2011: 42-hour average response, 23% never, 7x qualify if you try within an hour vs an hour later.2
- Invoca: 27% of home-services calls unanswered; <3% of voicemail-routed callers leave a message; 62% of buyers call before purchase.3,4
- Hatch: 60% average response on HVAC estimate follow-up in 163k campaigns.12
- Simpro 2025: 69% of trade owners say AI's biggest hit is workflow, not a new top of funnel.5
Run the one-week phone audit. If first human contact is over 5 minutes on paid leads, you do not have a lead-gen problem. You have a response problem. Fix that with the smallest tool that texts back and hands you the wheel.
FAQ
First: answer or text every inbound in under a minute, 24/7, then put a human on the hot ones. Second: follow up every unsold estimate on a sequence, not a sticky note. Third: ask for the Google review when the tech is still in the driveway. Those three raise booked jobs from the marketing you already run.
Yes. Jobber Receptionist, ServiceTitan's voice agent, and Podium's AI Employee all pick up, plus standalone voice vendors. The shop question is whether the bot can intake HVAC (no-cool vs maintenance), check service area, escalate emergencies, and write the job into the board you dispatch from.
A layer that takes the ring when the office cannot: after hours, overflow, lunch, storm days. Good ones qualify, book or hold, and alert on-call. Bad ones are IVRs with a nicer voice. Pair with missed-call text-back so a hang-up still becomes a thread.
No. The tools in this guide replace hold music and forgotten callbacks. Simpro's 2025 owner survey put AI's main impact on workflow optimization. Someone still has to recover the refrigerant, diagnose the board, and stand in the attic.
Not for lead gen. Use a writer for the eighth service page after the phones are caught. Jasper-class tools do not sit on your DID.
Sources
Every number on this page has a named source and a description of what it measured. None of these is an HVAC close rate unless the writeup says so.
- 1MIT / InsideSales lead response management (2007)
15,000+ B2B web leads, 100,000+ contact attempts, 6 companies. Contact odds drop 100x and qualification odds drop 21x when first contact moves from 5 minutes to 30 minutes. Contact and qualification, not HVAC close rate.
- 2Harvard Business Review, “The Short Life of Online Sales Leads” (2011)
Oldroyd, McElheran, Elkington. Test web leads to 2,241 U.S. companies: 42-hour average response among firms that responded within 30 days; 23% never responded. Companion 1.25 million leads: contact within an hour nearly 7x as likely to qualify vs one more hour, more than 60x vs 24 hours.
- 3Invoca, missed sales calls in home services
2022 Buyer Experience Report cited in their home-services missed-call writeup: 27% of home-services inbound unanswered; 62% of buyers call before they purchase; 76% stop after one bad experience; over 60% will pay more for good service.
- 4Invoca, turning missed sales calls into revenue
Platform data: fewer than 3% of callers pushed to voicemail leave a message.
- 5Simpro 2025 Trades Outlook
600+ trade and field-service businesses. 69% of owners see AI's biggest impact in workflow optimization, not in magically creating demand.
- 7Jobber AI Receptionist
Product: answers calls and texts 24/7, can offer to book, capture the request, or take a message, and can text back callers who hang up. Included on Plus; add-on on Grow.
- 8Plumbing & Mechanical, Jobber launches AI-powered receptionist (August 2025)
General availability writeup: 200,000+ conversations at launch. Jobber-cited stat that 96% of home service consumers say response time influences the purchase.
- 9ServiceTitan success story: Gulfshore AI voice agent
Named shop example. Finn answered a heat-wave call on a 10+ year system that was freezing up, booked same-day 2–5 p.m., and the tech closed $1,185. Finn had been live about two weeks.
- 10ServiceTitan, Gulfshore “job of the future”
Same story set attributes a $370,000 revenue lift over 30 days to Marketing Pro campaigns, which is a different product. Do not mash those two numbers into “the voice bot made $370k.”
- 11ServiceTitan Voice Agent product page
Vendor figures: ~70% booking including capacity conflicts and ~90% when capacity is not the blocker. Books against Adaptive Capacity.
- 12Hatch, HVAC estimate follow-up response rates
163,000 HVAC estimate follow-up campaigns, 2-day trigger after the quote. Average SMS/email response 60%. Best campaign 90.06% with 7 messages over 7 days (5 texts, 2 emails, 2 pause days). 1,229 campaigns at 80%+ response. Worst 17.82%.
- 13Hatch, Rescue Air testimonial
Vendor case study: 10% close-rate lift, 65% estimate follow-up response, and $260k won through rehash in the first two weeks. Your pile will not match theirs.
- 14Podium home services
Vendor customer quote: Premier Air's owner said converted leads went from 12% to 34% after Podium AI. HVAC names on the page include Frontier Air, Premier Air Colorado, Pride Air, Southern Heating & Cooling. Not a controlled study.
- 15Uncited “85% never call back” blog
The 85% figure that vendors paste on landing pages traces to an uncited 2016 blog post, not a study. Do not plan a shop on it.
- 16Uncited missed-call statistics report
Companion page repeating the same uncited 85% claim. Reject it.
- 19MIT study PDF (HubSpot-hosted copy)
Same 2007 MIT / InsideSales dataset as source 1: 5 minutes vs 30 minutes, contact 100x, qualify 21x.