If I advised a 6-person HVAC company, here's what I'd actually build with AI

I haven't done this work for an HVAC company yet. What follows is a framework: what I'd look at, what I'd build, what I'd skip, and what it would actually cost. If you run a trade business and this resonates, I'd like to hear from you.

I spend a lot of time thinking about where AI actually helps a small business versus where it's just a new way to spend money. The HVAC company is a useful test case because it's representative of most small service businesses: a handful of field techs, an office manager or owner handling scheduling and billing, tight margins, and a genuine problem with missed calls and follow-ups.

This post is my working theory on what I'd build. Not everything. The three things that would matter most.

The business I'm imagining

Six people: four techs in the field, one office manager, one owner who also runs calls
Serves a metro area, mostly residential, some light commercial
Gets 15 to 30 calls a day during season, 5 to 10 off-season
Uses some combination of: phone, text, Facebook, Google Business Messages
Scheduling is either in a paper calendar, a shared Google Calendar, or basic software like Housecall Pro or ServiceTitan
Invoicing happens after the job, sometimes days later
The owner answers calls after hours and on weekends

This is not a hypothetical perfect client. This is every HVAC company I've ever called.

Thing 1: AI intake and scheduling (not an AI receptionist)

The receptionist I built for Runari answers calls and captures info. For an HVAC company, I'd go further: the AI should qualify the call, check availability, and book the appointment.

What I'd build:

AI answers the call, asks what's wrong, captures address and contact info
Checks the calendar for next available slot
Offers two time windows
Books it and sends a confirmation text
Adds the job to the dispatch board with notes

What I'd use:

Retell or VAPI for the voice agent (~$0.09/min)
Calendly or SavvyCal for scheduling (~$15/month)
Zapier to connect them (~$20/month)
Existing CRM or just a shared Google Calendar

What it costs:

Voice agent: ~$50–$100/month for the missed and after-hours calls it catches
Scheduling tool: ~$15/month
Automation: ~$20/month
Total: ~$85–$135/month

What it replaces:

An answering service: $200–$500/month
Or the owner's evenings and weekends: harder to price

Why I'd build this first: Missed calls are lost revenue. A booked appointment is worth $150–$400 in HVAC. If this captures two extra appointments a week, it pays for itself in the first month.

What I'd watch out for:

Emergency calls need to route to a human immediately. The AI should recognize "no heat" in January and escalate, not offer a Thursday slot.
Existing customers with service agreements need different handling than new leads.
The calendar integration has to be real-time. A double-booked tech is worse than a missed call.

Thing 2: Post-job follow-up and review generation (not a chatbot)

Most HVAC companies finish the job, invoice, and disappear until the next breakdown. The smart ones follow up: "How was the service?" "Can you leave us a review?" "Your filter is due for replacement."

What I'd build:

24 hours after job completion, customer gets a text: "Hi [Name], [Tech] finished your [service] yesterday. How did everything go? Reply 1-5."
If they reply 4 or 5: "Glad to hear it. Would you mind leaving a quick review? [Google link]"
If they reply 1-3: "Sorry to hear that. [Owner] will call you within 2 hours."
90 days later: "Your HVAC filter is due for replacement. Reply BOOK to schedule."

What I'd use:

Simple SMS via Twilio (~$0.0075/message)
Zapier triggered by job completion in the CRM
Google review link (free)

What it costs:

SMS: ~$10–$20/month at 30 jobs/week, 3 messages each
Automation: included in the Zapier plan above
Total: ~$10–$20/month

What it replaces:

Nothing directly. This is new capability.

Why I'd build this second: Reviews are the cheapest customer acquisition in local service. A 4.8-star Google Business Profile with 200 reviews gets the call over a 3.2-star profile with 12 reviews. The follow-up also catches unhappy customers before they leave a bad review.

What I'd watch out for:

Don't automate the review ask for every job. Space it out. Google flags patterns.
The owner actually has to call back unhappy customers. If the "owner will call" message is a lie, this becomes a liability.
Filter reminders are useful but can feel spammy if overdone. Quarterly max.

Thing 3: Invoice automation (not AI, just plumbing)

This one has nothing to do with AI. It's just connecting the field work to the money.

What I'd build:

Tech marks job complete on their phone
Invoice generates automatically from the work order
Customer gets text: "Your invoice for $[amount] is ready. Pay here: [link]"
Payment lands in the business account
QuickBooks updates automatically

What I'd use:

Housecall Pro or ServiceTitan (if not already using): $79–$149/month
Or QuickBooks Online + Zapier + Stripe, at $30 + $20 + 2.9%

What it costs:

If already using field service software: $0 extra
If building from scratch: ~$50–$150/month

What it replaces:

Office manager doing data entry: 5–10 hours/week
Delayed invoicing: jobs billed days or weeks later, hurting cash flow

Why I'd build this third: Cash flow is the killer in small service businesses. The faster you invoice, the faster you get paid. Automated invoicing also means nothing gets forgotten. A $400 job that slips through the cracks is $400 you never see.

What I'd watch out for:

Techs need to actually mark jobs complete. If the culture is "I'll do the paperwork later," this breaks.
Some customers want paper invoices. Have a fallback.
The payment link needs to work on a phone. Most customers will pay from the text message.

What I'd skip (for now)

AI chatbot on the website. Most HVAC websites get 50–200 visitors a month. A chatbot that answers "Do you service my area?" is not worth the setup. A clear service area page and a phone number work better.

Predictive maintenance AI. The idea is appealing: analyze sensor data and predict failures before they happen. Reality: most residential HVAC systems don't have sensors, and the ROI on a $5,000 predictive system for $150 service calls doesn't math out.

AI dispatch optimization. Route optimization is a real problem, but Google Maps + a human who knows the territory is usually good enough for 4 techs. Save this for 15+ techs.

What this would cost, all-in

Build	Monthly cost	Setup effort
AI intake + scheduling	$85–$135	1–2 days
Post-job follow-up SMS	$10–$20	2–4 hours
Invoice automation	$50–$150 (or $0 if already using)	1 day
Total	$145–$305/month	2–4 days

For a company doing $40K–$80K/month in revenue, this is 0.2% to 0.8% of revenue. The question isn't whether you can afford it. It's whether the captured calls, faster payments, and better reviews return more than $305/month.

They almost certainly do.

The honest part

I haven't built this stack for an HVAC company. I've built pieces of it for myself and thought through the rest. What I don't know:

Whether Housecall Pro's API is reliable enough for real-time scheduling
How often customers actually reply to SMS follow-ups (my guess: 30–50%, but I don't have data)
Whether techs will consistently mark jobs complete in the field

What I do know: the problems are real, the tools exist, and the math works. The gap is usually execution, not technology.

Bottom line

If I were advising a 6-person HVAC company, I'd start with three things: capture more calls, get more reviews, and invoice faster. Two of those use AI. One is just good plumbing. All three are boring. All three work.

The mistake most businesses make is starting with the tool, "We should get AI," instead of the problem. The right sequence is: identify the leak, find the cheapest fix, build it, measure it, move to the next one.

If you run a trade business and this resonates, I do strategy roadmaps and implementation work: $3,000 for a phased six-month plan with real numbers and prioritized builds, or implementation from $5,000 with a scoped deliverable and fixed fee.

Either way, start with the problem. The tools are already good enough.