A pipe bursts at 9am. The homeowner calls three plumbers in a row. The first one to actually respond gets the job. I mapped where those calls go during a typical plumber's day — the answer is mostly voicemail. Here's what an AI-driven response setup actually looks like, and what the missed-lead math says about why it matters.
It's 10am on a Tuesday. You're two hours into a water heater replacement on Maple Street. The work is going smoothly — you're on schedule, the client's happy, you'll be done by noon. Your phone rings. Unknown number. You can't answer it — you're elbow-deep in the old unit and your hands are wet. You'll call back in an hour.
By the time you call back, at 11:15, the person on the other end has already booked a different plumber. Their kitchen was flooding. They needed someone in the next 30 minutes. You didn't answer. Someone else did.
That's the core emergency call problem in plumbing. Not that emergencies happen. Not that they're inconvenient. But that the window is measured in minutes, not hours. The homeowner calls the first plumber. If that plumber doesn't answer — and physically, you can't because you're on another job — they call the next one. By the time you call back, they're done deciding.
I spent time mapping where those emergency calls actually go in a typical plumbing operation. The answer is mostly voicemail, and the missed-lead math is uncomfortable.
Here's the scenario I traced across a few plumbing shops: A homeowner has an active problem. Toilet running, pipe spraying, water dripping through a ceiling. It's urgent. They call three plumbers.
Call 1: Goes to voicemail. They leave a message describing the problem.
Call 2: Also voicemail. They're getting frustrated. They text this one instead of leaving a message.
Call 3: A live human answers immediately. "I can be there in 45 minutes." Done. Plumber #3 gets the job.
From the homeowner's perspective, plumbers 1 and 2 aren't available. Doesn't matter that they're better or cheaper or closer. They weren't there. Plumber 3 was.
The problem is that plumbers 1 and 2 have 24-hour emergency lines in theory. In practice, they're the same person answering from inside a basement doing a complicated job. A live answer in 45 seconds is impossible.
The intervention isn't a complicated one. It's a two-question system that separates actual emergencies from scheduled work, and urgent emergencies from ones that can wait.
Question 1: "Is this an emergency that needs someone today, or are you looking to schedule a service visit?"
This question does the heavy lifting. A lot of calls that come through as "emergency" are actually just people wanting an appointment. They say "I have a plumbing problem" and expect emergency dispatch. When the AI asks, they clarify: "Oh, I need someone next week for an estimate." That's a completely different call. It goes into your appointment queue. The real emergency escalates.
Question 2 (if emergency): "Is water actively leaking right now, or did it stop and you want to prevent it from happening again?"
This separates true emergencies from urgent-but-not-critical ones. An active leak in progress is different from a slow drip that needs attention. A failed water heater that stopped heating is different from a pipe you know is cracked. The AI captures the context, and now your dispatcher has information. They know whether this is a "interrupt a job" emergency or a "first available" urgent.
Two questions. That's the entire AI component. It reads the homeowner's answers and routes accordingly.
The routing piece is where it gets practical. The AI doesn't try to book the appointment or negotiate a price. That's still a human operation. What it does is give your dispatcher real information:
"Active leak, pipe under sink. Homeowner's name: Sarah. Phone: [number]. Time pressure: yes."
Now your dispatcher can decide: Do we pull someone off the current job? Do we call the on-call plumber? Do we give this to whoever finishes their Maple Street job first? The decision is informed instead of blind.
The other thing the AI does is acknowledge the homeowner. They didn't get left in voicemail purgatory. They got answered. An actual voice (albeit not human at first) confirmed the problem, asked good follow-up questions, and is routing them to someone who can help. That alone changes the perception from "nobody answered my call" to "I got help."
My initial AI setup didn't have the triage questions. It was just: call comes in, automated message says "a technician will call you back within 30 minutes," and forwards to the dispatcher. Sounds efficient, right?
The problem: it fired on every missed call. Wrong numbers, spam calls, calls meant for another plumber — all of them got the "we'll call you back" message. The dispatcher would then spend time calling back people who were never legitimate customers. It created noise, not signal.
After the first plumber told me the system was generating more frustration than help, I added the two-question gate. Suddenly the dispatcher only gets routed actual customer calls that have been qualified. The noise disappeared. The conversion went up.
There's one edge case worth mentioning because shops worry about it: the homeowner who cries emergency but isn't really.
"My toilet's been running for two days but I'm calling now and saying it's an emergency because I want someone today."
The two-question system doesn't completely solve this, but it helps. When the AI asks "Is water actively leaking right now?" the homeowner either admits it's not an emergency or they double down on the lie. If they double down, your dispatcher still has the information that something's off — the story doesn't quite match. They can ask one more follow-up before deciding to interrupt someone's job.
It's not perfect. But it's better than the current system, which is "call and hope someone answers immediately."
There's a second layer to this that doesn't get attention but should. After the emergency call is handled and the plumber finishes the job, there's a window when the homeowner is most receptive to future planning.
Three days after the emergency call, an AI sends a follow-up: "We got your emergency fixed. While we were there, we noticed [specific thing — water stain under the sink, old water heater, corroded fitting]. This could become a bigger problem if we don't address it. Want to schedule a preventative visit?"
The homeowner is still in "I just had a crisis" mindset. The prospect of preventing the next one is immediately appealing. Conversion rates on this kind of follow-up are high — you're selling prevention during the exact moment when prevention feels necessary.
Let's do the math on a typical plumbing shop: one truck, one plumber, doing 4-5 jobs a day at an average of $300-500 per call. If the shop is missing 2-3 emergency calls per week to competitors — because they didn't answer and someone else did — that's $600-1,500 in lost revenue per week. Over a year, that's $30K-$80K in emergency work that went to someone else.
The two-question AI system costs basically nothing to set up. A Twilio phone number, some conditional logic, and three minutes of your time writing the two questions. The payback on recovering even one emergency call per week — $300-500 per call, 50 weeks a year — is immediate.
The other benefit is that homeowners who get through and get routed to you, even after a wait, are more satisfied than homeowners who got someone else because you didn't answer. You weren't the faster option, but you were clearly available and responsive. That drives repeats and referrals in a way that actually answering the phone would — but from a plumbing position perspective, that's impossible.
If you're running a plumbing operation and you're losing emergency calls, don't start by redesigning your whole workflow. Start small:
1. Set up a dedicated emergency number (separate from your regular line). This can be a Twilio number or a Google Voice number, forwarded to your dispatcher or on-call person.
2. Have the voicemail message say: "If this is an active emergency, press 1. If you're looking to schedule a service call, press 2." Two options. No AI required. Just voicemail routing.
3. For option 1 (active emergency), the message says: "We have an emergency plumber available. Please describe the problem after the beep." Let them leave a detailed message.
4. Your dispatcher listens to that message and makes the routing decision in real time.
That's the MVP version. No automation. Just better triage. Once you see how it changes your emergency response, you can add the AI layer that makes it smarter and faster.
The homeowner who's standing in a flooded kitchen doesn't care if a robot asks the triage questions or a dispatcher does. They care that someone answered, understood the problem, and told them when help is coming. A simple routing system that separates real emergencies from scheduled appointments does that. Everything else is just optimization.
The emergency call you miss is the one you never get back. The one you answer — even if you can't personally respond right away — is the one you still have a chance with.