Six sections covering what's actually broken in service businesses, how to size your specific leak in dollars, and how to recover it onto your calendar without replacing your CRM, your dispatcher, or anyone in your shop.
Or: what I learned watching my dad lose money on ads that were working.
About a year ago I was home in Saudi visiting my parents. My dad runs an orthodontic clinic. He's been in the same clinic since he finished his studies in France. Twenty-plus years.
He'd just started running Instagram ads. Hired a marketing guy who would film short videos with him, post them, run the paid promotion, the whole thing. So I asked him how it was going.
He said it was fine but he wasn't really happy with it.
I thought maybe the ads weren't performing, so I called the marketing guy. The marketing guy said something I keep coming back to. He said the ads were performing really well. The leads were coming in. The problem was the close rate was basically zero, and it was really frustrating to work with my dad.
I asked him what he meant. He said my dad would only reply to Instagram DMs the day after at the minimum, sometimes later. By the time he replied, the prospect had already gone somewhere else, or just forgotten.
So the marketing guy was doing his job. The algorithm was doing its job. The ads were doing their job. The whole funnel was breaking on the part where my dad had to actually answer the messages. And he wasn't doing it, because there were 40+ unread DMs sitting in his account every time he opened Instagram, and at least once a week someone was messaging him asking if he was based in Tripoli, Libya. The clinic is in Beirut. Different country. The Instagram targeting was loose enough that he was getting messages from people who couldn't possibly be his patients.
He was tired of it. So he stopped checking.
He was paying real money every month for ads that worked perfectly until they hit the part where a human had to be on the other side fast enough.
I told him I could automate it. Build him a system that replied to every new DM within 60 seconds, qualified the lead, filtered out the wrong-country messages, booked the legitimate ones. His exact reaction was, "that's exactly what I need."
I built it. Within a week his DMs were having real conversations. People were booking. The change wasn't dramatic in some movie way. The change was that he stopped opening Instagram and seeing 40 unread notifications. People started feeling like they got the attention they wanted, so they started trusting him, so they started booking.
That's the moment I understood what AI is actually good at when you use it right. It's not a gimmick. It's the thing that does the boring work humans can't keep up with.
Because every HVAC owner I've talked to has the same problem my dad had, just with different details.
The phone rings during a service call and goes to voicemail. The form submission lands at 9pm and nobody sees it until morning. The lead from Angi gets answered four hours late and they've already booked someone else. The big estimate gets sent and never followed up on. The customer who paid for an AC tune-up two years ago is forgotten until they buy from a competitor.
These are all the same problem. They're all the boring work nobody wants to do, and they're all where the money leaks out.
You're not competing against silence. You're competing against the next ad in the prospect's feed, the next contractor in their saved tabs, the next phone number in their Google search. Whoever responds first wins. And if you don't respond, you didn't lose a lead. You paid for a lead that you handed to your competitor.
That's what I want to talk about in this document. Not "more leads." Not "better marketing." The thing that's actually broken in most service businesses, and how to fix it without replacing your CRM, your dispatcher, or anyone on your team.
It doesn't matter how strong the strong parts are. The chain breaks at the weakest link. That's the whole game.
Most HVAC owners I look at are spending real money on ads, marketing, branding, vehicle wraps, websites. The strong parts of their funnel are very strong. The weak part is almost always the same one: what happens between "interested prospect" and "booked job." That's where the money disappears.
The reason it disappears is not that anyone is bad at their job. It's that the work in that gap is boring. It's repetitive. It's "reply to a DM, again, again, again, non-stop, and then check the form submissions, and then call the people who didn't pick up, and then follow up on the quotes from last week, and then send the review request to the customer from yesterday."
Nobody got into HVAC to do that work. People got into HVAC because they like fixing things, or running a crew, or being their own boss. The boring work is what makes the business actually run, and it's the first thing that gets dropped when there's a real problem to solve in the field.
This is true in every service business. My dad knew exactly what he needed to do with his Instagram DMs. He just didn't do it, because it was boring, took too long, and there was always a real patient in front of him who needed his attention more.
Someone who's passionate about workouts opens a gym thinking they'll spend their days coaching. They end up doing 95% sales, marketing, and admin. They barely have time for their own workouts. Same thing.
In most HVAC and home service businesses, the broken parts of the funnel look like this:
The dormant customer base sits in the CRM and nobody talks to it. Years of past customers, all of whom have aging equipment, none of whom are getting reached out to until they buy from someone else.
Every one of these is a leak. Every one of them caps how much money the business can make, no matter how much you spend on marketing.
You can spend $20,000 a month on Google Ads, and if 27% of your inbound calls go to voicemail during summer peaks, you've just lit $5,400 of that ad spend on fire every month. The math doesn't care that you have a great brand or a great team. It just measures the leak.
I'm not going to pitch you anything in this document. I'm going to walk through what I've learned about how service businesses actually leak revenue, and what to do about it.
The next sections cover:
You can read it and build it yourself. You can read it and decide it's not for you. You can read it and hire someone to do it. I don't care which one you do.
What I do care about is that you stop assuming your problem is "more leads." It probably isn't.
If you've ever looked at your monthly P&L and thought "we did the work, where did the margin go," there's a very good chance the margin went into the gap between when leads showed up and when somebody got back to them.
Let's get into it.
Or: where the money actually goes when you can't account for it.
Most HVAC owners I've talked to think their problem is "more leads." It almost never is.
The real problem is that revenue is leaking out of six specific places in the operation, all at the same time, every month. None of them are dramatic. None of them break the business overnight. They just slowly drip money into the floor while everyone is busy doing the actual work.
The leaks below are ranked roughly by how much money they cost a typical $1M-$5M HVAC contractor per year. Your business will be different. The point isn't that all six apply equally. The point is that almost every shop has at least three of them running at full leak rate and doesn't know it.
Let's go through them.
The phone rings. Nobody picks up.
Industry average is that 27% of inbound calls to HVAC companies go unanswered. During peak summer or peak winter, the number is higher. After hours, it's basically 100%.
Most owners hear "27%" and assume their shop is better. The honest answer is, they don't know. Out of 100 HVAC owners, maybe 10 actually track their missed-call rate as a real number. The rest are running on vibes.
Here's what missed calls actually look like from the customer's side.
When I was in college, I locked myself out of my dorm late one night. Called a locksmith. They didn't pick up. I waited maybe three minutes, then called a second one. The second locksmith's tech texted me back fast, said he'd be there in an hour. About an hour into all this, the first locksmith finally called me back. I told them I'd already booked. They lost a $200 job to a sixty-minute response gap.
The first contractor wasn't bad. They just weren't fast enough. And the customer never went back to find out why they didn't pick up the first time. They just moved on.
For an HVAC shop doing $1M-$5M a year, this leak alone is usually $80K to $400K of annualized lost revenue. The math is brutal in its simplicity:
The fix is an AI voice agent that picks up before the third ring, asks the basic qualifying questions, books legitimate jobs into the calendar, and texts the customer a confirmation. Not a "press 1 for service" tree. A real conversation. We'll cover the architecture in a later section.
Until then, the relevant point is this: the leak is real, the leak is measurable, and almost no contractor is measuring it. If you don't see the leak, you don't fix it.
This is the missed-call leak's bigger, sneakier cousin.
These are the prospects who didn't get ignored on the phone. They submitted a form on your website, sent a message through Angi, asked a question on your Facebook ad, DM'd you on Instagram, hit you up on Google's Local Service Ads. They're more interested than the cold callers, because they took the extra step to write something.
And then they wait.
Most contractors take 2 to 4 hours to respond to a web form or DM. Some take a full day. By the time the reply goes out, the prospect has either booked someone else, gotten distracted, or just forgotten they ever contacted you.
The data on this is consistent across industries: a lead contacted within 5 minutes converts at roughly 21x the rate of a lead contacted after 30 minutes. That number isn't because fast responses are magical. It's because every minute that passes is another minute the prospect gets distracted, sees a competitor's ad, considers another option, or just forgets.
My dad's clinic had a mom message him asking about a free consultation for her kid. He said yes. Two days later, she replied apologizing. She'd already taken her kid to a different orthodontist.
The clinic didn't lose because it said no. The clinic lost because the world kept moving.
I do this too. I once messaged two clothing stores with the same question. The faster one got my money. I wasn't loyal to either of them. I was loyal to whoever replied first.
Most owners think they're doing okay on response time. When you ask them, they'll say something like "we're pretty fast, usually within an hour or two." The customer's threshold is 20 minutes. That gap between perception and reality is where the leak lives.
There's an underrated point inside this leak that almost no agency talks about: missed calls are the visible problem, but slow DM and form response is the bigger one in dollars. The reason is volume. You probably get 10x more form submissions and DMs than calls. Even if your DM response gap is "only" 4 hours instead of 4 days, the total dollars walking out the door from slow response is bigger than the total from missed phone calls. Most contractors don't see this because the missed-call problem feels more visible. Phones make noise. DMs don't.
The fix is automated first-touch within 60 seconds across every inbound channel, with real qualifying questions, not "thanks for your inquiry, someone will reach out soon." A bot that says "thanks" is still slow response with extra steps.
This is the biggest leak in dollars for most HVAC shops, and almost nobody works on it.
Every HVAC company has a CRM full of people who used to be customers. Housecall Pro, Jobber, ServiceTitan, whatever you use, the names are in there. For a shop that's been running for 5+ years, that's typically 3,000 to 15,000 customers.
Most of them haven't been contacted in 12+ months. They're dormant. The shop technically still has them as customers, but operationally they've drifted off. They're not getting maintenance reminders. They're not getting check-ins. They're not getting offers. They're just sitting in the database.
When their AC eventually breaks or their furnace stops working, they Google "HVAC near me" and book whoever shows up first. The fact that you serviced their unit three years ago is irrelevant if they don't remember your name.
A while back I went to a barber who was a bit further from where I was living, but the cut was great. I walked out feeling fresh. I would have gone back to him for years. He never texted me. I never went back.
We both lost. He lost a recurring customer. I lost a good barber. The only reason it ended was that nobody bridged the gap.
Multiply that by every HVAC customer in every dormant CRM in America and you're looking at billions of dollars in lifetime revenue that just evaporates.
The math on dormant customer reactivation is genuinely insane. Reactivating a former customer costs about 1/5 the cost of acquiring a new one. Dormant customers convert at 2-3x the rate of cold leads, because they already trust you. Annual reactivation rates of 4-6% are realistic with a basic outreach system.
Here's why almost no HVAC shop does this: it feels weird. Most owners genuinely don't want to "spam" their old customers. They feel like reaching out is desperate. They'd rather just hope people come back on their own.
The reframe that fixes this: generic outreach IS spam, and you're right to feel weird about that. But specific outreach isn't.
The first one feels like spam because it is spam. The second one feels useful because it's specific to that customer's actual equipment, in their actual region, at the actual time it matters.
The fix here is a system that pulls equipment data from your CRM, knows when each unit was installed, knows what's likely going wrong at this stage of its life, and sends specific outreach at the right moment. That's not "automation" in the boring sense. That's done-for-you institutional memory.
Big estimate gets sent. Customer goes silent. Most contractors do nothing.
Industry data: 60-70% of estimates over $5K get exactly zero follow-up touches after the initial send. The contractor sends the quote, the customer doesn't reply within a few days, and the contractor mentally writes it off as "they decided no."
Most of the time, the customer didn't decide no. The customer is thinking, comparing other quotes, waiting for their spouse to look at it, or just got busy. A specific follow-up at the right moment closes a meaningful chunk of these.
I haven't personally sold a $5K HVAC estimate. But I know how I behave when someone sends me a big quote for anything: I sit on it, get distracted, and forget. If they followed up 3 days later with a specific question about my situation, that would actually feel useful, not pushy. Most people are like this.
The reason contractors don't follow up isn't pride. It's the boring-task pattern again. "Follow up on the quote from last Tuesday" is exactly the kind of repetitive, low-emotional-stakes work that gets dropped the moment a real problem shows up at the shop. It's the same reason my dad's Instagram DMs went unanswered for two days. Not because he didn't care. Because there was always a real patient sitting in front of him who needed his attention more.
That's $84K of revenue from work that was already 90% sold. The customer let you in their house, you walked them through the quote, you sent it. The work is mostly done. The follow-up is the last 10% of the work and where the money lives.
A customer has a great experience. They'd happily leave a 5-star Google review. Nobody asks them. The review never happens.
Most HVAC shops are getting 1-3 Google reviews a month when they could be getting 15-30 with a basic system. That's not just an ego problem. Google's local ranking algorithm uses review velocity (how many you get per month) as a major signal. Shops with consistent review flow rank higher in local searches, get more inbound calls, and feed the rest of the funnel.
This leak compounds. Fewer reviews leads to lower search ranking, which leads to fewer leads, which leads to smaller business, which leads to less ability to invest in operations, which leads to fewer reviews.
The fix is a 3-step ask: in the moment (right after the tech finishes the job), via SMS the same evening, and one polite reminder 48 hours later. In-the-moment asks convert at maybe 30-40%. Same ask 3 days later by email converts at 2-3%. Timing is everything.
There's no real psychological block here. Owners just don't have a system. The tech finishes the job, drives off, and nobody texts the customer. Build the system, the reviews come.
HVAC shops sell annual maintenance plans. Customer pays $200/year, gets two tune-ups. Plan expires. 30-50% of those plans don't renew, not because the customer was unhappy, but because nobody called them about it.
This is the gym membership pattern in reverse. Most people drop subscriptions just because nobody reminded them. I've done this with a gym membership before. Wasn't because the gym was bad. Was because nobody bridged the gap from "still subscribed" to "renewing actively." Same thing happens with HVAC service plans.
This number is smaller than the other leaks in absolute terms, but it's the highest-leverage one because it's recurring revenue. A single fix here doesn't add $30K once. It adds $30K every year, compounding.
Service contracts also tend to predict whoever buys an HVAC system replacement five years from now. Letting a contract churn isn't just losing $200, it's losing a future $8,000 system replacement to a competitor.
Read all six again and notice the pattern.
None of them are big problems. None of them break the business in a single week. Each one is a quiet, repetitive, boring task that doesn't get done because there's always something more urgent in front of the owner.
The work to fix them is exactly the kind of work humans don't keep up with reliably, especially when running a real business with real customers and real fires to put out.
This is what AI is actually good at. Not replacing your team. Not "transforming your business." Doing the boring work that keeps disappearing because nobody has time for it.
Or: do the math first, build second.
Most contractors who hear about operational automation react in one of two ways. Either they get excited and want to fix everything at once. Or they get skeptical and assume it doesn't apply to their shop. Both reactions are wrong, and both come from the same place: not knowing how big your specific leak actually is.
The leak is real. We covered that in Section 2. But "real" doesn't tell you anything actionable. You need a number.
Without a number, you can't decide whether this is worth doing. You can't decide which leak to fix first. You can't decide if it's worth $1,500 a month or $500 a month or zero. You can't even tell, six months from now, whether the fix actually worked.
The good news: sizing your own leak is genuinely simple. Most of the math is multiplication. The data you need is either already in your CRM, or it's a guess that's good enough.
Don't make this harder than it is. The goal of this exercise isn't precision. It's order of magnitude. You don't need to know whether your missed-call leak is $87,432 or $93,118 a year. You need to know whether it's $5K, $50K, or $500K. That's it.
Here's how to do it.
You have two options for sizing your leak.
Most people start with the calculator and then come back to this section if they want to dig deeper or sanity-check the result. Either order is fine.
The rest of this section is the manual version, leak by leak.
You don't need all of them. You need rough estimates for whatever you can find. If you don't have data, guess conservatively (low end). It's better to under-estimate the leak than to feel like the math was inflated.
What you need:
Take 15 minutes. Get the rough numbers. Now we calculate.
Where monthly missed calls = monthly call volume × missed-call rate. The 30% recovery rate is what an AI voice agent realistically converts from missed callers into actual booked jobs. (Real-world AI voice systems land in the 30-40% range; we use 30% to stay honest.)
That number is what's recoverable, not what's "leaking." There's a difference. The total leak (every missed call × full ticket × close rate) is bigger, but most of those callers were never going to book in the first place. The 30% recovery rate is the honest portion that gets converted by an actual response system.
Even at 30%, that's a number that justifies most operational fixes by itself.
If your recoverable number on this leak comes out under $30K/year, your missed-call problem is small (good for you, rare). If it lands between $30K and $200K, you're typical. If it's over $200K, this is your single highest-priority fix.
The math is harder here because "slow response" doesn't lose 100% of the lead, it loses a fraction proportional to delay.
The 30% comes from a conservative interpretation of the 21x conversion data. If a 5-minute response converts at 21x the rate of a 30-minute response, then any response slower than 30 minutes is operating at maybe 30% of its potential. The 70% gap is your leak.
Note: I said in Section 2 that the slow-response leak is often bigger than the missed-call leak in dollars. That's because written-channel volume is usually 5-10x higher than phone-call volume. If your form/DM volume is much higher than your call volume, expect this number to be larger than Leak 1.
This is usually the biggest leak in absolute dollars and the easiest to size.
The 5% reactivation rate is conservative. Real systems land between 4-8% annually. We use 5% to stay honest.
This is the leak that catches most contractors off guard. Six thousand customers in a CRM is normal for a 5-year-old shop. Five thousand of them being dormant is also normal. And 5% of them being reactivatable through specific outreach is real.
The 7% recovery rate is the honest one. Sequence-based follow-up systems typically recover 7-15% of otherwise-ghosted estimates depending on how good the sequence is and how long after the quote it fires. We use the bottom of that range to keep the math defensible.
That's $84K of revenue from work that was already 90% sold (the customer let you in their house, you walked them through the quote, you sent it). The work is mostly done. The follow-up is the last 10% of the work and where the money lives.
This one's the fuzziest of the six because it operates through SEO compounding rather than direct conversion. I'll show the math but flag that the dollar figure is more directional than precise.
The $400 per review comes from rough industry data: each additional Google review for a local service business contributes $300-500 in long-term lead value, weighted by review velocity. Not precise but directionally right.
I want to be honest: this number is wider-error-bar than the others. The actual SEO compounding might give you half that or twice that depending on your market and how saturated your competitors are with reviews. Treat it as "real but fuzzy" rather than "precise."
Healthy contract churn is around 10% (some customers move, some die, some buy a new house). Above 10% is the recoverable portion.
This number is smaller than the other leaks in absolute terms, but it's the highest-leverage one because it's recurring revenue. A single fix here doesn't add $30K once. It adds $30K every year, compounding.
Service contracts also tend to predict whoever buys an HVAC system replacement five years from now. Letting a contract churn isn't just losing $200, it's losing a future $8,000 system replacement to a competitor.
Add them up.
That's your total annual recoverable revenue.
For a typical $1M-$3M HVAC contractor, the honest total usually lands somewhere between $300,000 and $1,500,000 a year. If your number came out lower, either you're an unusually well-operated shop or you under-estimated. If it came out higher, you're typical for a busy summer season but the average across the year will be lower.
Now look at the breakdown. Which leak is the biggest? Which one would be easiest to fix?
In most shops, the priority order ends up being something like:
That's a rough ranking. Yours will be different based on which leaks are biggest in your specific business.
The numbers you just calculated are already the recoverable portion, not the raw leak. The full leak (every missed call × full ticket × close rate, every ghosted estimate, etc.) is bigger. What you calculated is the conservative, defensible piece that a well-built operations system can actually move onto your calendar.
If the totals still feel insane, two things are probably true at once.
Right that there's more in your funnel than you thought. Most contractors don't believe the math the first time. They see $400,000 a year recoverable from a $2M business and assume the formulas are broken. They aren't. The math is conservative on every leak (30% missed-call recovery, 7% quote recovery, 4% dormant DB reactivation). Everything is on the low end of what's plausible.
Wrong that all of it lands in month one. Real-world delivery curves take 60-90 days to ramp. Month 1 of a fix usually delivers 20-30% of the steady-state number. By month 3, you should be hitting close to the calculated total. By month 6, you're at or above it.
Either way, write your number down. Save it.
Or: why this isn't a chatbot, an answering service, or any of the other things you've already been pitched.
If you've been running an HVAC business for more than a year, you've probably been pitched some version of this six different times by six different people. Each one had a different name for it. Each one had a slightly different angle. None of them quite landed.
You've been pitched chatbots that sit on your website and reply to questions. You've been pitched AI voice agents that pick up the phone after hours. You've been pitched answering services with humans on the other end. You've been pitched lead generation agencies that send you Facebook ads and "warm leads." You've been pitched marketing automation tools that send emails on a schedule. You've been pitched "AI for your business" by guys who couldn't explain what their product actually did.
After enough of these pitches blur together, the natural conclusion is that none of it works. That's a reasonable conclusion based on the specific things you've been pitched. It's also wrong, because none of those things were actually trying to fix the problem we've been talking about for the last three sections.
What I want to do in this section is define what an AI Operations System actually is, name what it isn't, and explain why the distinction matters for the math we just walked through.
That's not marketing language. That's the most honest description of the category. The work it does is the same work a really good operations manager would do for you, if you could find one, afford one, and trust them to do it without forgetting. It picks up calls when nobody else can. It replies to forms in 60 seconds. It remembers every customer who hasn't been touched in 18 months. It follows up on every estimate without being asked. It asks for the review at the right moment. It calls the customer whose service plan is about to expire.
Three things define the category, and these three things are what separate it from everything else you've been pitched.
It sits on top of what you already have. It doesn't replace your CRM. It doesn't replace your dispatcher. It doesn't replace your phone system. It plugs into all of them and runs the work between them. If you switch CRMs next year, the system keeps running. If your dispatcher quits, the system keeps running. The operational layer is independent of any specific tool you use.
It runs continuously across every channel. Phone calls, web forms, DMs, LSA leads, missed calls, dormant customer outreach, quote follow-up, review requests, contract renewals. Not one channel at a time. Not on a schedule. All of it, all the time. The reason most operational fixes fail is that they only handle one channel, which means the leak just shifts to whichever channel the fix doesn't cover.
It does the boring work. This is the real category boundary. Everything an AI Operations System does is something a human could theoretically do, and something a human won't reliably do. Reply to a DM at 9pm. Call a dormant customer about their aging unit. Send a follow-up to a $7K estimate three days after it was sent. These are tasks that don't get done because nobody wants to do them, and the second a real customer walks in, they get dropped. The system does them anyway, every single time, without getting tired, distracted, or annoyed.
The clearer way to understand the category is to look at what it gets confused with.
I want to address something honestly because the word "AI" is doing a lot of marketing work right now and most of it is dishonest.
AI is not magic. It's not going to "transform your business." It's not going to think for you. It's not going to replace your judgment. Anyone selling you AI as a transformation strategy is selling you something that doesn't exist.
What AI is actually good at is exactly the kind of work we've been describing. It's good at responding fast across multiple channels at the same time. It's good at remembering everything without forgetting. It's good at executing repetitive contextual tasks without getting bored. It's good at scaling without getting more expensive.
You could hire a team of people to manually pick up every missed call, reply to every DM in 60 seconds, follow up on every quote, and call every dormant customer. The math just didn't work. Even at minimum wage, that team would cost more than the recoverable revenue. The economics required either dropping the work (which is what most contractors do) or finding labor cheap enough to make it work (which doesn't exist).
AI changed the unit economics. The work that previously required eight people now runs on infrastructure that costs a few hundred dollars a month. The boring work finally became affordable to actually do.
That's the whole reason this category exists right now and didn't exist five years ago. Not because AI is magic. Because the math finally works.
If you take one thing from this section, take this:
The people who would handle the boring work that actually moves revenue if you could find them, afford them, and trust them to do it.
You've never been able to hire that team because the math doesn't work for an HVAC contractor doing $1M to $5M a year. Hiring three operations people at $40K each is $120K of overhead that pays for itself only if the work they do generates more than $120K of additional revenue, and most owners can't take that bet on three new hires they don't yet trust.
The Operations System makes that math work. The recoverable revenue we walked through in Section 3 (probably $300K to $1.5M a year for your shop) goes from "theoretical money" to "money your operations team is actively recovering, every day, without you having to hire them or manage them or trust them."
That's what you're actually buying when you buy an AI Operations System. Not chatbots. Not voice agents. Not marketing automation. The operations team you couldn't afford to hire, running continuously, doing the work that nobody on your current team has time to do.
The next section walks through how this is actually built. The specific tools, how they connect, why this stack and not another. If you don't care about the technical architecture, you can skip Section 5 and go straight to Section 6, which is the build-versus-buy decision framework.
If you care about how it works under the hood, keep reading.
Or: how this is actually built, named by name, with no hand-waving.
Most agencies hide the architecture. They don't want you to know which tools they're using because if you knew, you could either build it yourself or hire someone cheaper to build it for you. That's a fair business move on their part.
I'm not going to do that.
Two reasons. First, the value isn't in the tools. The tools are commodities. The value is in knowing what to do with them, in what order, with what specific configuration, for HVAC specifically. Anybody with two months of focus and a high pain tolerance can wire up n8n and Twilio. Doing it well, doing it for HVAC specifically, doing it so the system actually delivers 30+ jobs in 60 days, that's the actual hard part.
Second, you should never trust a system you don't understand. If a contractor signs with us and has no idea what's running under the hood, they have no way to verify the system works, no way to evaluate alternatives, no way to keep operating if we ever stopped supporting them. That's a bad deal for the contractor. So we show the architecture and explain the parts.
If you read this section and decide "I can build this myself," that's a valid outcome. You probably can if you're technical and have 2-3 months to focus on it. If you read it and decide "I want someone else to handle this," that's also valid. The point is making an informed choice.
The full system has five layers. Each one does one job. Each one can be swapped out for an alternative if you ever need to. None of them replace your existing tools, they sit on top of and around them.
n8n is the engine that ties everything together. It's the thing that says "when X happens, do Y, then Z, then check if A, and if so do B."
If you've heard of Zapier or Make, n8n is the same kind of tool but built for serious work. Zapier is fine for hobbyist automations. n8n is what you use when you have 50 of those running simultaneously, with branching logic, conditional follow-ups, error handling, and complex data transformations between steps.
The reason we use n8n specifically: it's open source and self-hostable (we run it on infrastructure we control), it handles complex workflows without per-step pricing that breaks at scale, and it has every integration we need (CRM, Twilio, Retell, Supabase, Slack, Gmail) as native nodes.
Twilio handles the physical phone call. When somebody dials your business number, the call is routed through Twilio before it reaches a human (or in this case, an AI).
You've never heard of Twilio, but you've definitely interacted with it. When Uber sends you a "your driver is 3 minutes away" text, that goes through Twilio. When DoorDash calls to confirm a delivery, that goes through Twilio. They're the infrastructure layer of about 30% of SMS and voice traffic on the internet.
For our purposes: Twilio gives us a phone number customers can call, and lets us programmatically answer those calls and route them to the AI voice agent (Retell) instead of voicemail. A normal phone line can't do that.
Retell is the AI that actually talks to your customer on the phone. When the missed call gets picked up, Retell is the voice on the other end.
Honest about what it does and doesn't do: It does sound human (real customers in real tests don't reliably identify it as AI). It does follow scripts with flexibility (we define the conversation flow but it handles natural deviations). It does NOT replace your judgment. If a customer needs something complex, the AI hands off to a human. The AI is good at standard inbound (~90% of calls). The other 10% need a human, and the system knows it.
We use Retell because voice AI is a moving target. Six months from now there might be a better voice provider. The system is built so that if a better provider emerges, we can swap it without rebuilding the rest.
Supabase is where the system remembers things.
Every conversation, every quote, every customer interaction, every dormant customer's status, every review request sent, every contract renewal due. All of it lives in Supabase. Think of it as the system's brain.
Why we use Supabase rather than just storing everything in your CRM: your CRM stores what jobs were done. The Operations System needs to track much more than that. Trying to shove all of it into your CRM would clutter it beyond usability and tie us to that specific CRM forever. Supabase keeps the operational data separate, clean, and CRM-agnostic.
The fifth layer is your existing CRM. Housecall Pro, Jobber, ServiceTitan, whichever one you already use.
We don't replace it. We don't migrate data out of it. We don't ask you to switch.
What we do is connect to it via API so the Operations System can read and write the right things at the right time. From your dispatcher's perspective, nothing changes. They keep using the CRM they're used to. New appointments show up. New customers get added. The Operations System is invisible to them, except that suddenly there's more work on the schedule than there used to be.
This is the part most agencies get wrong. They try to replace your CRM. Keeping your existing CRM as the source of truth is non-negotiable for a system that's actually meant to last.
Here's how the five layers actually work together in a real situation. This is the missed-call scenario, end to end.
Normally: the call goes to voicemail, the customer hangs up, they call your competitor.
Instead, here's what happens:
A few common alternatives and why we don't use them.
One more alternative worth addressing:
You could. A skilled developer could build the whole system from scratch in 3-6 months for $50K-$150K depending on the city. There's nothing in our system that requires a special algorithm or proprietary technology. The reason most contractors don't go this route is that 3-6 months and six figures is a lot of risk for a system that may or may not work for your specific business.
Using established tools (n8n, Twilio, Retell, Supabase) is faster, cheaper, and more reliable than building everything custom. The thing worth building custom is the configuration, not the underlying infrastructure.
A core principle of the system is that no component is locked in.
If a better voice AI than Retell emerges in 12 months, we swap it. If you decide to switch CRMs from Housecall Pro to ServiceTitan, the system follows you. If Twilio raises prices and a competitor like Vonage becomes cheaper, we move. If n8n ever became unavailable (extremely unlikely, it's open source), we could rebuild the workflows in any equivalent orchestrator.
This matters because the technology landscape will absolutely change in the next 5 years. Voice AI will get better. New CRMs will emerge. Phone infrastructure will evolve. A system that's locked into one specific vendor for any of these layers becomes obsolete the moment that vendor stops innovating. Modular architecture is the only way to build something that lasts.
The configuration and the workflows (which are what we actually charge for) are portable. They live in n8n as exported JSON files. If you ever needed to take the system somewhere else, those files come with you.
If you've read this far, you now know more about the technical architecture of an AI Operations System than 90% of HVAC contractors who buy one.
That's not flattery. It's the point.
Most agencies don't show this much because they're afraid you'll figure out you can build it yourself. I'm showing it because if you decide to build it yourself, that's a fair outcome. The decision tree in Section 6 will help you figure out whether building it yourself is actually a good use of your time, or whether having someone else handle it is the better deal.
For now, the relevant point is: this isn't magic. It's not proprietary. It's not patented. It's well-understood software components, configured carefully for HVAC operations, running on infrastructure that costs a few hundred dollars a month. The hard part is the configuration and the operational discipline to run it well. The components themselves are commodities.
Or: now that you know what's broken and how it works, what should you actually do?
You've read about the leaks. You've seen the math. You know what an AI Operations System is and what it isn't. You know how it's built.
Now you have to decide what to do with that knowledge. There are exactly three honest paths.
You can build it yourself. You can hire someone (us or someone else) to build it for you. Or you can do nothing and let the leaks keep running.
All three are valid. The right one for you depends on five specific factors I'll walk through. The point of this section is to give you a real framework for figuring out which path fits your situation, not to push you toward whichever path makes me money.
By the end of this section, you should know which one you're picking and why. If the answer is "I'm doing nothing," that's fine. If the answer is "I'm building this myself," that's fine too. If the answer is "I want someone else to handle this," we're here for that.
Let's go through it.
The decision isn't about whether AI Operations Systems work. It's about whether building one yourself is a better use of your time and money than hiring it out, or whether either of those is better than doing nothing.
These are the five factors that actually matter.
How much time do you have? Not "how much time do you wish you had." How much time can you genuinely commit to this in the next 90 days, on top of running your business?
Building yourself: 200 to 400 hours of focused work for somebody who's already technical. Double that if you're not. That's two to four months of evenings and weekends.
Hiring: 5 to 10 hours total. Onboarding calls, giving access to your CRM and phone system, reviewing the configuration. The rest is somebody else's problem.
Doing nothing: zero time, you keep losing whatever your calculator number was per year.
How technical are you? Be honest. There's no shame in either answer.
If you can read a YAML config file, deploy a Docker container, write a basic shell script, and debug an API integration when it breaks at 2am, you have the technical baseline. Probably half the people reading this don't, and that's fine.
The honest middle ground some contractors land in: they have a son, nephew, or friend who's technical and would help for free or cheap. If that person exists and is genuinely available for 200 hours of focused work, building yourself becomes much more realistic.
How much capital can you put at risk?
Building yourself: $300-600/month in tool subscriptions once running. Total upfront cash: maybe $1,000-3,000 if you do all the work.
Hiring it built: ranges wildly. Mid-tier US agencies charge $5,000-15,000 setup plus $1,500-3,000 monthly. Specialist firms charge $20K-50K setup. Generic "AI agencies" with no HVAC experience charge whatever they think you'll pay.
Doing nothing: whatever your calculator number was. If it said $400K/year recoverable, doing nothing is a $400K/year decision. Most contractors don't think about it that way because the cost is invisible. It's still real.
How much do you actually trust outsiders with your business? This matters more than most contractors admit upfront.
The Operations System touches almost everything: your phone system, your CRM, your customer database, your booking flow. If you hire someone to build it, you're giving them deep access to the operational core of your business.
If you've been burned by vendors before, micromanage anything that touches your customer data, or genuinely believe nobody understands your business as well as you do (sometimes true), building it yourself or hiring a single trusted contractor is probably better than hiring an agency.
If you're happy to delegate when the deliverable is well-defined and the SLA is clear, hiring out is faster and lower-stress.
How urgent is the leak?
If your business is doing fine and you have time to figure this out at your own pace, urgency is low. Building yourself works because you don't need this running by next month.
If your business is stalling because of operational gaps right now, summer is coming and you can't keep up with call volume, or your competitors are eating your lunch with faster response times, urgency is high. Hiring out makes more sense because you don't have time to learn this from scratch.
If your business is generally healthy but you've identified a specific recoverable opportunity (your dormant database is worth $200K and just sitting there), urgency is medium. Either path works depending on the other four factors.
Run yourself through this and see where you land.
Right path if: The recoverable revenue from your calculator is genuinely small (under $50K/year). Your business is at capacity already and you couldn't handle the new jobs anyway. You're planning to sell or wind down the business in the next 12 months. The leaks aren't actually a constraint on your growth right now.
Let me say something honest about that third option. "Do nothing" is a real and valid choice. Most contractors who hear about Operations Systems end up doing nothing and their businesses are mostly fine. Not great, not growing as fast as they could, but fine. If you're in a market with limited demand or you genuinely don't want to grow more, the leaks aren't really leaks. They're a mild inefficiency on a business that's where you want it to be.
But if you do want to grow, "do nothing" is the same as "I'm okay losing the calculator number every year." Make sure you're picking that path actively, not by default.
If you're leaning toward building yourself, here's the realistic timeline based on what we've actually built.
The hardest part isn't any single component. It's the integration work and the configuration tuning. Each individual piece is well-documented. Making them all work together for HVAC specifically, without breaking anything when your CRM updates or Retell pushes a new version, is where most DIY builds get stuck.
If you're leaning toward hiring, here's what a good engagement looks like, regardless of whether it's us or somebody else.
This is the path most contractors take, so let me be honest about what it looks like.
You finish reading this playbook. You think "interesting, I should look into this." You don't. The next time you think about it is six weeks from now when you have a particularly bad week of missed calls. You think "I should really look into that thing." You don't.
A year from now your business is where it is. Not bigger, not smaller. The leaks are still running. The recoverable revenue is still recoverable, just by your competitor instead of by you.
That's not catastrophic. Most businesses run this way. You'll be fine.
The fact that you're still here means something in the math woke you up. The honest question is whether you act on it now or six months from now after losing another six months of recoverable revenue.
If you've read this far and you're leaning toward "I want someone else to handle this," we're here for that.
We work with one HVAC contractor per metro. We charge $5,000 setup and $2,500/month. We guarantee 30+ booked jobs by Day 60 or full refund, no fine print. We don't lock you into long contracts. The architecture is the one we walked through in Section 5, configured for your specific business.
If we're not a fit, we'll tell you on the diagnostic call. If your business is too small, we'll tell you. If you'd be better served building it yourself, we'll point you to resources for that.
The diagnostic call is 30 minutes. We walk through your leaks, size them honestly, and tell you whether the math justifies our offer for your specific business. Most calls end with us saying "here's what we'd do, here's what it would cost, here's what you'd recover." Some end with us saying "you don't need us, here's what to do instead."
Run the calculator first. If the recoverable number isn't enough to justify $2,500/month, the diagnostic call is a waste of both our time. Calculator is on the homepage, takes 60 seconds, no email required.
If the number is real, book the call.
Whatever path you pick, pick it actively, not by default.
If you're building, set a start date this week. Block out the first weekend. Order whatever hardware or accounts you need. Don't let "I should learn more first" become a year of inaction. You've already learned more than enough to start.
If you're hiring, run the calculator, book the diagnostic, and get on a call this week. The offer either works or it doesn't, but the only way to know is to actually have the conversation.
If you're doing nothing, decide that consciously. Write it down somewhere. "I read the Opsmatics playbook on [date], decided my situation doesn't justify acting on it, and chose to keep the operations the same." Six months from now, when you're frustrated with another bad call volume month, look at what you wrote and decide if you still believe it.
The whole point of this document was to give you enough information to actually decide.
Decide.
That's the playbook. If you got value from it, the calculator is on the homepage and the diagnostic call is one click away. If you didn't, no hard feelings, that's why we wrote it. Either way, thanks for reading.
Jad Domloge, founder of Opsmatics