AI receptionists: scope, cost and when to use one
What an AI receptionist can handle, where a person takes over and how to assess cost, illustrated with a plumbing-company workflow.
An AI receptionist is one possible way to handle incoming enquiries. The right scope depends on the questions callers ask and what happens next. This guide uses a plumbing-company workflow as a concrete example; it is not a restriction on who I work with.
An AI receptionist for a plumbing company answers the phone on the first ring, every time, including at 2 a.m. It finds out three things: what the job is, where it is, and how urgent. Then it texts those three things to a human, who calls back within minutes and closes the job themselves. It does not book. It does not dispatch. It does not pretend to be a person. That is the whole job, and it's enough.
One call, the way it should run
Here is a call the way I'd want it to go. This is an illustration I wrote for this post, not a recording. The script for my first client was written from real call transcripts, but those aren't mine to publish.
The phone rings once.
Assistant: "Thanks for calling [Company]. I'm the phone assistant, not a person. I can take the details and get a plumber to call you straight back. What's going on?"
Caller: "The kitchen tap won't shut off. It's running into the sink, so it's not flooding, but I can't stop it."
Assistant: "Got it, a tap that won't shut off. What's the address?"
Caller: "14 Elm Street, the blue house."
Assistant: "14 Elm Street. Is it causing damage right now, or can it wait an hour or two?"
Caller: "It can wait a bit. I've put a bucket under it."
Assistant: "Okay. I'm sending this to [Owner] now. Expect a call back within a few minutes. Is this the best number to reach you on?"
Caller: "Yes."
Assistant: "Thanks. Someone will call you shortly."
The text lands on your phone a few seconds later:
New call, 14:07. Kitchen tap won't shut off, not flooding. 14 Elm Street, blue house. Can wait an hour or two. Wants a callback on this number.
You read it under the van door. You call back four minutes later and quote it from your head, the way you always have. The assistant asked three questions and got out of the way. That's the entire product.
Now the other version. Same greeting, but the caller says "I can smell gas" or "there's water coming through the ceiling."
Assistant: "That sounds like an emergency. If you smell gas, go outside now and call the gas emergency number. It's [the number you gave me at setup]. I'm alerting [Owner] as urgent right now. Are you somewhere safe?"
That's it. One instruction, no questions about which floor the flat is on. Your text is flagged urgent and comes with a ring, not a buzz. For water everywhere, it tells them where the stopcock usually is (under the kitchen sink in most UK homes, at the main shut-off or the meter in the US), takes the address, and marks it urgent. Three sentences, then hand off. The worse the call, the less the assistant should talk.
What it must not do, and why
Three things I've deliberately left out. People ask for all three, and I say no to all three.
It doesn't book. The assistant has no idea where your crews are. It doesn't know the Tuesday 10 a.m. slot in your calendar is a boiler swap that's already running long. A booking made without that knowledge is a promise you then have to break, and the caller hears it as your promise, not the software's. The human who calls back knows the board. Let them book.
It doesn't promise times. "Someone will be there by three" is the sentence that gets you a one-star review. The assistant says "a plumber will call you back within a few minutes," because that's a promise it can keep and I can measure. The arrival time is yours to give, once you've looked at the day.
It doesn't pretend to be a person. It says it's an assistant in the first sentence. Partly because a caught lie costs more than the sale, and in a town with ten plumbers the story travels. Partly because most people are fine with it. When Housecall Pro asked 1,040 US homeowners in October 2025, 53% said they were comfortable with AI handling initial inquiries. The ones who aren't comfortable get a human calling back in minutes anyway. Nobody's fooled and nobody needs to be.
If a vendor tells you their receptionist can book straight into your calendar, ask them one question: how does it know where the van is. Listen carefully to the answer.
What it costs
Two prices to know, and then the honest comparison.
Mine: builds start at $4,500, fixed scope, fixed price, quoted before I start, so your number doesn't move once it's agreed. Then a monthly fee for the phone line, the minutes and the monitoring, quoted with the build. I won't put a monthly figure here because it depends on your call volume and how much watching the thing needs in the first months.
Here's what the alternatives publish on their own pricing pages, as of the day I wrote this:
| Option | Who answers | Published price |
|---|---|---|
| Smith.ai, human-first | US receptionists, 24/7 | $300 a month for 30 calls, $11.50 a call after that |
| Smith.ai, AI receptionist | AI first, human on transfer | Free for 25 calls a month, then $3.00 a call; Pro tier $150 a month |
| Ruby | US virtual receptionists | $250 a month for 50 minutes, up to $1,725 for 500 minutes |
| Goodcall | AI agent, set up yourself | $79 a month per agent, unlimited minutes, 100 unique customers then $0.50 each |
Now a worked example. Say your phone takes 90 calls a month, three a day, five minutes each. On Smith.ai's human-first Basic plan that's $810 for the month. On Ruby you'd need 450 minutes, which puts you on the 500-minute plan at $1,725. On Smith.ai's AI tier, 25 calls are free and the other 65 cost $3.00 each, call it $195. On Goodcall's Starter you'd pay $79, assuming those 90 calls came from fewer than 100 different people.
So why would anyone pay $4,500 up front for mine? Two reasons, and if neither applies to you, buy the cheap one. First, the script. The self-serve tools give you a web form where you type what the bot should say on a Sunday evening. Mine is written from transcripts of your actual calls, so it asks about the airlock, not "your plumbing needs." Second, someone is answerable. When it mishears an address at 11 p.m., you have my number, not a support ticket.
The human services are the real comparison, not the bots. A human receptionist is better than my assistant at soothing an angry caller. My assistant is better at answering the ninth ring at 2 a.m. on a Saturday without a per-minute meter running. Pick the one that matches when your calls come in.
Whichever you pick, do the missed-call maths first. Run your own numbers before you talk to anyone, including me.
When not to buy one
I recommend against this more often than you'd think. Four cases where you should keep your money:
- Your phone rings twice a day and you answer both. There's nothing to catch. An answering system for a phone that gets answered is a monthly fee for nothing.
- You already have a dispatcher or an office manager who answers. They know the board, they know the customers, and they book. Don't put a machine in front of the best receptionist you'll ever have. If they go home at five and the phone doesn't, that's an after-hours conversation, not a replacement.
- You want it to book, and your software can't accept it. If the thing you actually want is auto-booking into Jobber or ServiceTitan, I'm the wrong person, and I'd argue the timing is wrong too. Fix the catch and the callback first. Booking is a different product with a different set of ways to go wrong.
- You sell mostly through tenders and contracts. If your work is commercial, arrives by email, and gets decided in a spreadsheet, your phone isn't the leak. Your quote follow-up probably is. Different post.
If one of those is you, close this tab. I'll still be here if the phone starts ringing more than you can answer.
The one plumbing company that has signed
One. Not twelve, not "dozens of trades businesses." One plumbing company has signed a contract for the system I've described here. It goes live in October 2026.
The script came from real call transcripts, not from imagining what a plumbing call sounds like. That's the part I'm proudest of, and also the part that took longest, because real callers don't say "I have a plumbing emergency." They say "it's making a noise" and "the thing under the sink."
There are no results yet. I haven't invented any, and I won't. The first 30 days go up on the built page when they exist, whether they're impressive or boring. If the assistant mishears three addresses in the first week, that goes up too. In a niche this small, the honest number is worth more than the flattering one.
If you want the longer version of what I build for plumbing companies, it's on the plumbing page.
Where to start
Don't start with a demo. Start with a count. For one week, write down every call you missed and every voicemail that never called back. Most owners guess low. Then put the number into the calculator with your average job value and see what it says. If the number is small, you've saved yourself a monthly fee. If it's ugly, book a growth call and I'll look at your phone, your quotes and your follow-up with you. Thirty minutes, free, and I'll tell you if the answer is no.
Questions people ask
How much does an AI receptionist cost for a plumbing company?
Off-the-shelf tools run from free to a few hundred dollars a month: Smith.ai's AI tier starts at $0 for 25 calls, Goodcall at $79 a month per agent, on their pricing pages in September 2026. Human answering services like Ruby start at $250 a month for 50 minutes. My builds start at $4,500 fixed, then a monthly fee for the line, minutes and monitoring, quoted with the build.
Does it work with ServiceTitan, Housecall Pro or Jobber?
It hands every lead to a person by text, so it works next to whatever software you run. Pushing the lead into ServiceTitan, Housecall Pro or Jobber as a new customer or job request is scoped per build, because each one accepts data differently. It never auto-books into any of them.
Can it book appointments?
No, by design. It doesn't know where your crews are, how long the current job is running, or which slots are already spoken for. It gets the job, the address and the urgency, texts you, and you book from your head when you call back. Booking is a separate product with a separate set of problems.
What if it can't understand the caller?
It asks once more, in plainer words. If it still can't get the answer, it stops trying and hands off to a human with whatever it has: usually the caller's number, sometimes the address, and a note that it didn't catch the rest. A confused caller gets a person, not a third attempt.
Will customers know it's an AI?
Yes. It says so in the first sentence, before the caller has said a word. Nobody is tricked, and the person who calls back is a real plumber. In Housecall Pro's October 2025 survey of 1,040 US homeowners, 53% said they were comfortable with AI handling initial inquiries. The rest get a human in minutes anyway.
Sources
- 1.Housecall Pro: The New Home Service Standard, homeowner survey (October 2025) · 1,040 US homeowners surveyed 23 October 2025; 53% comfortable with AI handling initial inquiries
- 2.Smith.ai: human-first receptionist pricing · Starter $300/mo for 30 calls, $11.50 per call over; Basic $810/mo for 90 calls; Pro $2,100/mo for 300 calls
- 3.Smith.ai: AI receptionist pricing · Free tier 25 calls/mo then $3.00 per call; Pro $150/mo platform fee with per-call rate from $2.00; Enterprise $500/mo
- 4.Ruby: virtual receptionist pricing · Starter $250/mo for 50 minutes; Standard $395 for 100; Popular $720 for 200; Enterprise $1,725 for 500; no setup fees
- 5.Goodcall: pricing · Starter $79/mo per agent, unlimited minutes, 100 unique customers then $0.50 each; Growth $129; Scale $249
Written by Ivar André Knutsen
I build and run AI systems, internal tools and workflow automation. You work directly with me from the first conversation through implementation and support. About Ivar
Want this looked at in your business?
We look at where you want the business to go, what is slowing you down and where AI could make a useful difference. You get a clear recommendation: a tool to try, a focused automation, a broader system or a closer look at the process. Any build is scoped and quoted before work starts. Free, no obligation.
Single automations are quoted on the call: a setup fee plus a monthly retainer to run them. Full systems start at $4,500, fixed scope, fixed price.
Book a free 30-minute call