Every trades owner I talk to is getting sold AI right now. Call answering, dispatching, review management, website builders, follow-up bots. 10 pitches a month, all of them promising the same thing, and no way to tell which ones actually pay back.
I have had this conversation on the podcast with about numerous home service owners over the last two years. Some of them are running AI across the whole business and would not go back. One of them lost six figures finding out where it does not work.
This is what they actually said. Not what a software company says. What operators running HVAC, plumbing, electrical and multi-trade shops found out with their own money.
The $100,000 Lesson
Start with the failure, because it’s useful story and helpful when you consider adding AI to various areas of the business.
Ryan Williams bought a trade business and wanted to fix speed to lead. Reasonable goal. Calls were getting missed, response times were slow, and an AI answering service looked like the obvious fix. So he tested one.
Before the test, booking rate on calls answered by a human was running around 85 percent. With AI handling those calls, it never broke 40.
The reason matters more than the number. His customer base skews over 50 and Southern. Real accents, conversational nuance, people who want to explain the problem their own way before they get to the point. The AI could not handle it. Not because it was a bad product. Because it was the wrong product for that customer.
He ran it for several months before the pattern was clear enough to act on. The booking drop likely cost over $100,000 in revenue.
Here is the part most people skip when they tell this story. He did not throw the tool out. He cut it for live answering and kept it for chat and aggregator integration, where it dramatically improved response times. Same technology. Failed badly in one channel, worked well in another.
The lesson is not that AI is bad. It is that the same tool can fail in one place and win in another, and only your own numbers will tell you which.
Where It Is Actually Working
Plenty of it is working. Here is where operators told me they are getting real value, with names, because vague claims are what got everyone confused in the first place.
Call handling and speed to lead
Mark Paup runs Golden Rule, a multi-trade operation doing plumbing, HVAC and electrical. He is using AI across dispatching, call handling, after-hours booking, speed to lead, and back office work. He is also the most direct about what it takes to get there, which I will come back to.
Finding revenue in calls you already took
Josh Crouch built a company around this. He has analyzed hundreds of thousands of home service calls, and his finding is that the average booking rate is around 55 percent while most owners believe they are closer to 80. AI is what makes it possible to review that volume of calls and find the gap. A person listening to recordings one at a time never gets through enough of them to see a pattern.
That is the strongest version of the AI argument in my opinion. Not replacing the conversation. Reading all the conversations you already had and telling you what you missed.
Reporting and dashboards
Josh Kelly made the point that a dashboard that would have cost $150,000 to build a few years ago is now something one person can put together. That changes the math on financial visibility for a $10 million shop considerably.
Austin Blanchard said something similar about internal tools. His team can now connect data and build simple applications without a development team.
That workaround is one answer to weak native reporting. The full comparison of field service platforms covers when it makes more sense than switching.
Fleet and asset decisions
Renee Milum works in fleet, and she described using fleet data to improve replacement timing, spot maintenance patterns, and make cycling decisions. That is a cost center most shops manage by feel, and it is a good example of AI doing something a human could do in theory but never actually does.
Back office and admin
Brent Garrett uses it for emails, proposals, spreadsheets, process documentation and phone call analysis. Unglamorous, and probably the highest hit rate of anything on this list.
Marketing spend analysis
Sarah Ghirardo framed it as using AI to uncover the truth about where marketing money is going. Automating the analysis so you can eliminate waste rather than guessing at it.
Where It Keeps Failing
The failures cluster, and the pattern is worth knowing before you buy anything.
Brigham Dickinson has listened to thousands of real calls. His position is that AI can answer a call and book a job, but the jobs it books cancel at a higher rate when nobody follows up. It works as a first pass for straightforward calls, after hours and overflow. It does not work on the hard ones. The “what do you charge to come out” call, the “I just want a ballpark” call, the customer who wants to talk to a tech. Those still need a trained person.
Dan Antonelli was blunt about AI websites. His argument is that AI is a tool and not a brand strategy, and a site generated without real photography, local visibility and an actual point of view will cost you more in lead acquisition than it saved you in build cost.
Put those next to Ryan’s $100,000 and the pattern is clear. AI struggles where the interaction carries trust. It performs where the job is volume and pattern recognition. A homeowner deciding whether to let you into their house is the first category. Reading nine hundred call recordings is the second.
The Point Almost Nobody Makes
James Hatfield said the thing I keep coming back to. His line was that the AI gold rush is really a data gold rush.
His argument is that the contractors who capture and organize their conversations, their training, and their documentation end up with an asset that multiplies the value of the company. The tools will keep changing. The organized data is what compounds.
From where I sit, that has a direct financial version. If your books and your field software do not agree, AI does not fix it. It just gives you the wrong answer faster and with more confidence.
I open a lot of QuickBooks files where the ServiceTitan numbers and the ledger have not tied out in a year. Point an AI at that and it will produce a beautiful analysis of numbers that are wrong. Here is why those two systems drift apart and what it takes to bring them back together. Do that before you buy anything that promises insight.
The CFO Question: Does It Pay Back?
This is the part that is missing from almost every AI conversation I hear in this industry. Everyone debates whether the technology is good. Almost nobody runs the number.
Ryan Williams is the model here, and not because the test worked. Because he measured it. He knew his booking rate before, he watched it after, and when it did not recover he killed it in the channel where it was losing and kept it in the channel where it was winning. Most owners never set up the test at all. They buy the tool, it feels modern, and eighteen months later nobody can say whether it earned anything.
Four questions before you sign up for anything:
- What does it cost fully loaded? Not the monthly fee. The monthly fee plus the setup time plus the person who has to babysit it for the first ninety days. That third number is the one nobody quotes you.
- Which number is it supposed to move? Name it before you start. Booking rate. Average ticket. Days to collect. Hours of admin per week. If you cannot name one, you are buying a feeling.
- What is that improvement worth in dollars? Do the arithmetic first. If your booking rate goes from 55 to 60, that is a 9 percent increase in booked jobs, because you measure against the 55 you started from. On a $5 million book that is roughly $450,000 against marketing you already paid for. Now you know what the tool has to beat. What good actually looks like in a P&L tells you what that revenue is worth once it lands.
- What is the test period, and what result kills it? Decide the kill criteria before you start, while you are still objective. Once you have spent the money and told your team about it, you will find reasons to keep it.
That last one is where most of the money gets lost. Not on bad tools. On tools nobody was willing to turn off.
Mark Paup’s Warning
Of everyone I have talked to about this, Mark had the most practical thing to say, and it was not about the technology at all.
AI is not plug and play. It needs a strategy, a deliberate rollout, and a real conversation with your employees before you expect it to work. His team is getting value out of it across the business, and the reason is that they treated it like an operational change instead of a software purchase.
Most failed implementations I hear about were bought like software and never rolled out like a change. The tool showed up, nobody owned it, the team routed around it, and six months later it was an unused line item.
So Do You Need AI, Or Do You Need a CFO?
I get a version of this question a lot, and I will give you the honest answer rather than the self-serving one.
AI has gotten genuinely good at the mechanical work. Categorizing transactions, reconciling, summarizing, drafting, finding a pattern across a thousand records. A lot of what used to be junior finance work is going to get absorbed, and it should. That is not a bad thing.
Here is what it will not do. It will not tell you no. It does not know you are already carrying a note on the building. It does not know your partner wants out in three years and that this decision makes that harder. It does not know you tried this same thing in 2023 and it did not work. It has no memory of the payroll you sweated in February, so it cannot weigh how much risk you can actually stomach in October.
The other thing is more basic. AI is only as good as the books underneath it. Ask it to analyze a P&L where field labor is buried in overhead and maintenance agreement cash is booked as revenue on the day it is sold, and you will get a confident, well-written, completely wrong answer. Garbage in, eloquent garbage out.
So the order matters more than the choice. Get the books clean and closing on a schedule first. Here is the month-end close checklist I use. Then add the tools on top of numbers that are actually true, and you will get real leverage out of them.
Most owners at $3 million to $15 million do not need to choose. They need the reporting fixed and then they need to be smart about what they automate. If you want to talk through which is which in your shop, schedule a call and bring last month’s P&L.
Questions Owners Ask
Can AI replace my bookkeeper?
Increasingly it can replace a lot of the recording work, meaning categorization, matching and reconciliation. What it does not replace is somebody owning the close and catching the things that are wrong. If your bookkeeper is only doing data entry, that role is changing. If they are closing the month properly, they are doing something AI is not close to doing on its own.
Is AI call answering worth it for a home service company?
It depends on your customer, which is the honest answer nobody wants. One operator I had on the show saw booking rate fall from around 85 percent to under 40 with an older, rural customer base, and it cost him over $100,000 before he pulled it. The same tool worked well for him in chat. Test it in one channel, measure booking rate before and after, and decide the kill criteria before you start.
What should I automate first?
Back office work where nobody is talking to a customer. Proposal drafting, documentation, call analysis after the fact, spend analysis. High volume, low trust. Leave the first conversation with a homeowner alone until you have proof it holds up with your specific customer base.
How do I know if an AI tool is paying for itself?
Name the number it is supposed to move before you buy, measure it for ninety days, and compare the improvement in dollars to the fully loaded cost including setup time. If you cannot name the number, that is the answer.
