What an AI Receptionist Actually Costs a Small Business (and When It Pays for Itself)

The short answer, before anybody sells you anything

Most AI receptionists for a small business land somewhere between about $50 and $500 a month, plus a setup fee, plus per-minute charges once you pass the included minutes. That's the range I see quoted around Northern Colorado right now. But the price is not the decision.

The decision is one line of math: calls you miss in a month × the share you'd have closed × your average job value. If a roofer in Windsor misses 20 calls a month, closes one in four, and averages $9,000 a job, that's $45,000 of exposure. A $300 tool is not a real conversation at that point. If you're a shop that misses three calls a month at $80 a ticket, the honest answer is don't buy this yet.

Nobody selling you the software is going to run the second version of that math with you. So let's run it.

What these things actually cost, including the fees nobody quotes

Here's how the pricing usually breaks apart when you read the order form instead of the landing page:

• Monthly platform fee. The number on the pricing page. Usually includes a bucket of minutes.
• Overage per minute. This is the one that bites. Every call over your bucket bills by the minute, and callers who ramble bill the same as callers who book.
• Setup or onboarding. Sometimes waived, sometimes a one-time fee. Waived usually means an annual contract.
• Phone number and telephony pass-through. Pennies, but it's a line item.
• Integrations. Pushing a booking into your calendar or CRM is frequently a higher tier, not the base plan.
• Your time. The real cost. Somebody has to write what the thing says, list your service area, list what you don't do, and correct it for a few weeks.

My rule of thumb: take the quoted monthly number, add the overage you'll realistically hit, then add 20% for the oopsie fund. That's the true, true cost. If the vendor won't show you a sample bill from a shop your size, that's information too.

The one calculation, done properly

Get four numbers. Three of them you already have.

1. Missed calls per month. Not guesses. Pull your call log on your phone or in your carrier portal and count inbound calls with no answer, plus voicemails you returned more than an hour later. Late is missed.
2. Your close rate on calls you do answer. Most trades I sit with are somewhere between one in three and one in five on inbound. Use your own.
3. Average job value. Revenue divided by jobs, last 90 days.
4. Capture rate. Be conservative. An AI receptionist doesn't save every missed call, because some callers hang up on any robot. Assume it saves half.

Worked example, hypothetical: a clinic in Greeley misses 30 calls a month, closes 40% of the ones it answers, and averages $220 in first-visit revenue. Thirty missed × 50% captured × 40% closed × $220 = $1,320 a month recovered. Against a $250 all-in tool, that's an asset, not a liability.

Flip it. A specialty shop in Berthoud misses four calls a month at $95 average. Four × 50% × 30% × $95 = $57. Now you're paying $250 to recover $57. Don't buy it. Buy something else, or fix the phone habit.

I have to make you more than I cost you. Software should be held to the same standard.

What you're actually buying, and what a human still has to do

An AI receptionist is good at four jobs: answering on the first ring at 9 p.m., taking a name and number, answering questions you've written answers to, and booking a slot on a calendar. That's it. That's the product.

What it does not do is judgment. It doesn't know that the caller with the wet ceiling should jump the queue over the caller pricing a spring project. It doesn't know your crew is already double-booked Thursday because your foreman told you in the truck.

So you still owe it three things a human has to build: a service-area list, a we don't do that list, and an escalation rule that says which calls ring your cell right now. Skip those and you've bought a very polite voicemail with a subscription.

Where these things fall on their face

Test for these before you sign anything, because they are the failure modes I see over and over:

• Warm transfers. Half the fights are here. Can it actually reach you, and what happens when it can't? "I'll have someone call you back" with no ticket created is a lost job.
• Names and addresses. Colorado street names and last names get mangled. Make it read back and spell.
• Pricing questions. Callers ask what it costs. If your bot dodges, they hang up. Give it a range and what moves the number. Homeowners reward the one who shows a range.
• Angry callers. Somebody's basement is flooding at 6 a.m. A cheerful script makes that worse. There should be a keyword path straight to a human.
• Silence and background noise. Job-site calls, wind, a caller on speaker in a truck on I-25.

Call your own bot from a parking lot with the windows down before you turn it loose on your customers.

What we learned running Maddox on our own line

We don't recommend anything we haven't put on our own phone, so our receptionist, Maddox, answers the real Schreier Group number. Not a demo line. The one people actually call.

Three honest lessons.

One: the setup is writing, not technology. The hard part wasn't standing it up. It was writing by hand what we say, what we won't say, what we charge for, and where we refer people out. Every wrong answer it gave traced back to something I hadn't written down. That's not an AI problem. That's a business-clarity problem the phone happened to expose.

Two: it gets things wrong, and you have to read the transcripts. Every week for the first stretch. If nobody reviews the logs, you don't have a receptionist, you have a rumor.

Three: the win isn't the conversation, it's that somebody picked up. A caller who reaches a voice on the first ring stops dialing down the list. That's the whole ballgame for a crew on a roof in July or plowing a lot in January. The research everybody in home services quotes says the vast majority of companies fail to respond within five minutes, and inbound phone calls convert dramatically better than web forms. Answering is the differentiator. The clever script is a bonus.

We also run this stuff through the same wringer we use on vendor claims for clients in our hype audit. If it doesn't survive a real phone line, it doesn't ship.

Cheaper things to try first if your volume is low

Before you subscribe to anything, here's the zero-to-cheap ladder. Some of these cost zero American dollars.

• Turn on missed-call text-back. Your phone or CRM can auto-text "Sorry, on a job. Text me what you need and I'll reply within the hour." Texts get opened. This alone closes a chunk of the leak.
• Rewrite the voicemail greeting. Say your name, say when you call back, say to text instead. Thirty seconds of work.
• Forward to a second human. Spouse, part-timer, office manager, two hours a day during the crunch window.
• Traditional live answering service. Humans, per-minute, often competitive with the AI tiers for low volume.
• Put a real phone number and a real face on your website. Half the missed-call problem is people who called because they couldn't tell from your site whether you do their job.

How do you eat an elephant? One bite at a time. Fix the voicemail this week. Buy software next quarter, if the math still says so. If you want a second set of eyes on which bite is first, the free Business Checkup is built for exactly that.

How to run a 30-day test without signing a year

Here's the plan I'd give you Monday morning.

1. Week 0: Count your missed calls for one honest week. Multiply by four. Write down your close rate and average job. Now you have the ceiling on what this can be worth.
2. Pick month-to-month. Annual contracts exist because churn is high. Pay the setup fee to keep the exit open. Two paths to victory, and you want the one where you can leave.
3. Route only overflow at first. Rings your cell four times, then the bot. You keep the good calls, it catches the leakage. Lower the bar to entry.
4. Read every transcript for 30 days. Score three things: did it get the name and number right, did it book or escalate, would you have kept that customer?
5. Judge on one number. Booked jobs that came through the bot, times average job value, minus true cost including overage. Positive, keep it. Negative, cancel it and go back to text-back. No hard feelings.

That's it. No hype, no faith. Thirty days and a spreadsheet. Skepticism is a feature here, and if something on your phone line is already on fire rather than merely flammable, that's a triage conversation, not a software purchase.

Questions I get about this

Todd Schreier has owned a bakery, built escape rooms, rehabbed houses and sold businesses. He runs Schreier Group with his wife Amanda out of Johnstown, Colorado. Call the office any time: (970) 591-4263.