# AI receptionist for HVAC companies: what it does on a call Published: 2026-09-20 About the author: Eshan Cheema Tagged: AI voice agent, Contractors, ServiceTitan, Jobber Related: Missed-call revenue calculator and worksheet (https://plainbuilt.ai/guides/missed-call-revenue-calculator-worksheet) Numbers in this article are being confirmed. ## In three lines - An AI receptionist is the AI voice agent: it answers every call, asks the qualifying questions, and creates the job or the appointment in ServiceTitan, Jobber or Housecall Pro in real time as the call happens. - A call outside its scope is warm-transferred to a person with a summary: the caller's name, the issue, and the qualifying details. It is live in 3 to 4 weeks, after 5 business days in shadow mode. - The one voice-agent deployment number on this site is a concrete contractor's, not an HVAC company's: a 23% lift in monthly lead capture within 60 days of the voice agent going live, on ServiceTitan and Twilio. An AI receptionist for an HVAC company answers every call, at 2 a.m. and in the summer rush, asks the qualifying questions, and books the job into ServiceTitan, Jobber or Housecall Pro while the caller is on the line. A call outside its scope goes to a person with a summary. It is live in 3 to 4 weeks. ## What does an AI receptionist do on an HVAC company's line? Owners say receptionist. On this site it is the [AI voice agent](/services/ai-voice-agents), and it is the same thing: software that picks up the phone and holds a normal conversation, not an IVR that routes callers through menus. It is trained on your business before it takes a call: your services, your pricing tiers, your service areas, your booking rules, your after-hours rules, and your FAQs. On each call it works out what the caller needs, asks the qualifying questions, and creates the job or the appointment in your scheduling software in real time. The voice is picked to fit your brand, more direct for commercial work and friendlier for residential, and callers regularly do not realize they are speaking to an AI. The qualifying questions are the ones your CSRs ask from memory today, and week 1 is spent writing them down. The agent reads the customer record by phone number first, then the location, the job history, and the open windows, so it asks only what the record cannot answer: what is wrong, whether it is an emergency, and which window the caller wants. We train it on four kinds of edge case by name: emergency escalations, calls that cover more than one service, pricing questions, and scheduling conflicts. ## How does it book into ServiceTitan, Jobber or Housecall Pro? During the call, not after it. Jobs and appointments are created in ServiceTitan, Jobber or Housecall Pro in real time as the call happens, and on ServiceTitan the agent books directly to dispatch. HubSpot is a direct integration for teams that run their pipeline there, FieldEdge is also wired in, and the agent is deployed into your existing phone system through Twilio. The order of the reads and writes is on the [integration map](/guides/servicetitan-jobber-integration-map). The agent reads the open windows live, so it offers a time that is actually free. It writes once, last, after every slot the booking needs is filled and the caller has said yes: the appointment or the visit, with the job, the type, the window, and the call notes. Then it reads back what was booked, from the system, not from memory. A second write at call end puts the summary, the transcript link, and the recording where the office looks for them. The service page lists a 23% lift in monthly lead capture and a 93% call resolution rate as its measured results across deployments; the 23% matches the concrete contractor's number below, and the written commitment names the number for your operation. A call outside its scope hands off. Emergencies and anything the agent was not trained for are warm-transferred to your on-call person with a summary, the caller's name, the issue, and the qualifying details, so your CSR picks up with context instead of starting from zero. If the write to the scheduling system fails, the fallback is a person too. ## What does the agent say about recording, and what changes by state? Plainbuilt's own line says the call may be recorded and that the caller is talking to an AI agent, on every call. Whether a state requires the announcement is a separate question, and the answer changes at the state line. [Indiana](/locations/indiana) is one example: its wiretap article, IC 35-31.5-2-176, reaches only a person who is neither the sender nor the receiver of the communication, so a party to the call may record it, and a third party needs one party's consent. Have counsel read the section for your state before the agent takes its first call. The state changes one more setting. Indiana runs on two clocks, so a cutover written for Eastern time, say 5 p.m., starts the after-hours script an hour early on a Central-time shop around Evansville. The after-hours window is set by the county the business is in, not by the state. ## What is an after-hours HVAC call worth? Run the number on your own phone log first. Missed calls per month, times your average ticket, times your close rate, is the revenue that leaks each month. The [missed-call worksheet](/guides/missed-call-revenue-calculator-worksheet) runs it by hand, and the [calculator](/tools/roi-calculator) runs it with a 60% capture-rate assumption that is a working number, not a promise. On the calculator's starting numbers, 100 missed calls, a $500 ticket and a 40% close rate, the leakage is $20,000 a month and the recovery at 60% capture is $12,000 a month. Count 30 days before go-live and 30 days after, on the same metric, and the capture rate becomes something you measured. I tell owners the after-hours line is the first thing to fix, before the website and before the ad budget. 60% of service calls come outside business hours, the number the voice agents page carries, and the founder page has the line I wrote for it: a missed call at 9 p.m. is a lost $3,000 job. For an HVAC company the after-hours call is the one where the caller has already decided to spend; nobody phones at 9 p.m. to browse. The company whose line answers, asks the qualifying questions, and books the visit gets that job. The company whose line rings six times and goes to voicemail is the next number on the caller's list. That is why most clients start with a single module, and why it is typically the voice agent. ## What happens in the 3 to 4 weeks before go-live? Week 1 is the audit. On ServiceTitan we pull 30 days of call recordings and the dispatch board to map where leads leak; discovery takes 2 to 3 hours of your operations team's time. Weeks 2 and 3 are the training and the integration, side by side. Week 4 is shadow mode: on the [contractors agent](/products/industry-agents/contractors) the agent runs alongside your CSRs for 5 business days with every booking reviewed, then cutover, then weekly tuning calls for 60 days on real conversion data from your dispatch system. A complicated integration can add 1 to 2 weeks. Every deployment carries 30 days of post-launch support and a kill switch, and we track the workflow against its projection at 30, 60, and 90 days; if it underperforms by 20% or more, we redesign on our dime. Most contractor engagements start with the voice agent and add quote generation in month 2. ## What we saw in deployment There is no HVAC case study on this site, and I will not write one until there is. The scripted demo on the contractors page is a plumbing call with made-up names and numbers. The voice-agent deployment number we have is a contractor's, in a different trade. A regional concrete contractor in the Midwest ($10M to $15M in annual revenue, 60+ field crew, a 4-person office) estimated it was losing 15 to 20% of inbound leads to voicemail after hours and at peak call times; in summer, hold times stretched past 2 minutes. We deployed a voice agent into its existing phone system through Twilio, integrated with ServiceTitan dispatch: it handles after-hours calls, qualifies leads, books appointments directly on the dispatch board, and escalates emergencies to the on-call manager. Diagnosis ran in weeks 1 to 2 and architecture in weeks 3 to 4, then the voice agent was deployed and tested in production in weeks 5 to 8. The case study is dated January 15, 2026, and its measurement dates are still being confirmed. Its one voice-agent metric is monthly lead capture: from an estimated 15 to 20% of inbound leads lost to voicemail to +23%, measured within 60 days of voice agent deployment, as leads that previously went to voicemail were qualified and booked automatically. [Read the case study](/work/mccaslin). What went wrong: the case study's note is about the diagnosis, a week spent building the revenue leak map from scratch when the client's call tracking software already held the data; in contractor engagements since, we pull call recordings on day one, which is why the audit on the contractors page starts with 30 days of ServiceTitan call recordings and the dispatch board. [not yet published: Eshan confirms this article as his own words and its numbers] ## FAQ ### What happens to the calls the AI receptionist cannot handle? It hands off. Emergencies and anything outside its scope are warm-transferred to your on-call person with a summary of the caller's name, the issue, and the qualifying details, so your CSR picks up with context instead of starting from zero. If the write to your scheduling system fails, the fallback is a person too, and the agent says so rather than reading back a confirmation it does not have. ### Which HVAC scheduling software does it book into? ServiceTitan, Jobber and Housecall Pro: jobs and appointments are created there in real time as the call happens, and on ServiceTitan the agent books directly to dispatch. HubSpot is a direct integration for teams that run their pipeline there, FieldEdge is also wired in, and the agent is deployed into your existing phone system through Twilio. ### Does the agent tell callers the call may be recorded? Plainbuilt's own line does, on every call: it says the call may be recorded and that the caller is talking to an AI agent. Whether a state requires it is a separate question. Indiana, for one, is a one-party state under IC 35-31.5-2-176, so a party to the call may record it. Have counsel read the section for your state before the agent takes its first call.