What an AI Answering Service Costs, and When You Still Want a Person

A plumber with both hands inside a water heater doesn’t answer the phone. The call goes to voicemail, and the caller decides whether to leave a message or try the next name on the list.

An AI answering service takes those calls instead. It handles the routine part of the conversation and sends the owner a written summary when the call ends.

I should say up front that I run Rinqly, one of the companies in this market. So I’ve put competitors’ published prices right next to ours, all as of October 2026.

What an AI answering service does

On an ordinary day it answers questions from the facts you’ve given it, such as your hours, service area, prices and policies.

It takes messages, books and moves appointments in your calendar, and passes urgent calls to you or whoever is on duty. After every call you get a summary, usually with the transcript or a recording attached.

The caller explains the problem in their own words, and the AI talks it through with them. With a live answering service, trained people in a call center work from a script you approved. Some providers combine the two and hand the difficult calls to a human.

The software only knows what you tell it. Change your prices without updating its instructions and it will quote last month’s numbers with total confidence.

If you’re weighing the bigger phone-system vendors such as Nextiva and RingCentral, Vizologi’s roundup of AI answering machines covers them. I’m sticking to the services small businesses tend to buy on their own.

One call, start to finish

Say an HVAC company gets a call at 9:40 on a Saturday night. The office closed at five.

The call has to reach the AI first. The usual setup keeps the existing business number and adds forwarding at the phone carrier, either for every call or only for calls nobody picks up after a few rings. I’d start with the second. Your staff still answer when they can, and the AI only gets the calls they miss.

The caller says the furnace stopped, the house is getting cold, and there’s a newborn inside. Speech recognition turns that into text while they talk, and a language model checks it against the owner’s instructions, which spell out the services, the zip codes the company covers, what the AI may say about price and what counts as an emergency. Newer systems handle the listening and the talking in one speech-to-speech model, which shortens the gap before each reply.

A cold house with a newborn inside matches the emergency rule, so the AI transfers the call to the technician on call. If the caller only wants a routine tune-up, the AI offers two open time slots from the calendar and books one. Either way, the owner finds a short summary on their phone with the caller’s name, number and what they need.

Callers notice the pause

People leave very short gaps when they talk. A study published in PNAS recorded conversations in ten languages and found an average gap of about 200 milliseconds between one speaker finishing and the next one starting. The most common gap was zero (Stivers et al., 2009). The same researchers found that people deliver a real answer faster than a non-answer, and a “yes” faster than a “no.”

So a long pause on the phone sounds like hesitation. The tricky moment comes at the end of the caller’s sentence. People stop mid-thought to remember an address, and an AI that jumps into that silence cuts them off. Measuring any of this is harder than it looks. In September, our own call logs showed waits of 12 to 19 seconds on a call I had been on myself, and I hadn’t heard a single one. The log was timing from a signal our voice model never sends. I judge pauses from recordings now. For a vendor, that means calling their demo line yourself instead of trusting a recorded sample.

What it costs

You’ll run into three ways of billing.

Live answering services charge for the minutes a human receptionist spends on your calls. PATLive, a US service that answers only with people, lists a pay-as-you-go plan at $49 a month plus $2.99 a minute, and bundles from $99 for 50 minutes to $759 for 500 minutes.

Most AI services sell a monthly bundle of minutes and charge per minute after that. My AI Front Desk charges $99 for 200 minutes, then 25 cents a minute. Rinqly, my company, starts at $49 for 140 minutes, then 35 cents a minute.

A few AI services bill per caller. Goodcall charges $79 a month for up to 100 unique customers, then 79 cents for each additional person, and doesn’t count minutes at all.

The fourth option is hiring. The Bureau of Labor Statistics puts the median receptionist wage at $18.27 an hour, or $38,010 a year, as of May 2025. Benefits made up 31.5% of total compensation costs for full-time private-industry workers in June 2026, so a full-time receptionist costs an employer roughly $4,600 a month. That buys about 40 of the 168 hours in a week.

Pricing out one month of calls

PATLive calls its 200-minute plan the most popular and estimates it at about 80 calls a month. Here’s what that volume costs on each option’s published pricing, checked on October 9, 2026:

At this volume the AI services cost $250 to $280 a month less than the human service, and the gap grows with every extra minute, since PATLive’s extra minutes start at $1.79. A receptionist costs far more but also knows your regulars and does other work between calls.

Some vendors won’t show a price until you’ve told them your call volume, and even the published plans leave out the details that decide the bill. So before you sign, ask whether they bill by the second or round every call up to a full minute, and whether spam calls and wrong numbers come out of your plan. Ask what a minute costs once the bundle runs out, whether there’s a setup fee or a minimum term, and which features cost extra. Spanish, CRM sync and outbound calls often do.

For the per-call versus per-minute question, we’ve worked through the break-even math with Ruby’s and Smith.ai’s published plans in a separate guide to answering service pricing on our company blog.

Who gets the most out of it

It pays off fastest where the person who should answer the phone is busy doing the paid work: trades and field service, clinics, salons, small law and accounting offices.

Look at last month’s missed calls. If a lot of them came in the evenings and on weekends and most callers wanted the same few things (are you open, do you cover my area, what will it cost, when can you come), I’d try one, especially if a single booked job is worth more than a month of the service.

Speed is the other argument. In a study published in Harvard Business Review, researchers analyzed 1.25 million sales leads at 42 US companies. Companies that tried to reach a lead within an hour were nearly seven times as likely to qualify it as companies that tried an hour later, and more than 60 times as likely as those that waited 24 hours or longer. When the same team sent test inquiries to 2,241 US companies, 23% never responded at all (Oldroyd, McElheran and Elkington, 2011).

Those were web inquiries, and I wouldn’t stretch the exact multiples to phone calls. I just can’t see why a caller who hits voicemail would wait more patiently than someone who filled out a web form.

I’d still hand some calls to a person, starting with emergencies.

A medical question or a smell of gas needs a human or an emergency line fast, and the AI’s only job is to get the caller there. An angry customer calling about a botched job wants a person who can apologize and fix it. So does anyone negotiating a big job where price and scope shift mid-call. And some people just won’t talk to software. Give them an easy way out, either a transfer or a callback at a time you’ll actually keep.

A hybrid setup covers most of these. The AI takes the routine calls and passes anything outside its rules to a human. If you’d rather ease in, run it after hours only, then widen its schedule once the transcripts show it handles your callers well.

How to test one before you sign

Most of these services offer a free trial, so spend ten minutes calling in from your own phone:

1. Ask something that’s in your instructions, then something that isn’t. A good system admits it doesn’t know and takes a message instead of inventing an answer.
2. Spell out an email address and a street name, then check the summary. Spelled names, emails and addresses are hard for any speech system.
3. Interrupt it mid-sentence. It should stop and let you talk.
4. Ask for a human, and count the steps it takes to get one.
5. Book an appointment, call back to move it, then open your real calendar and check.
6. Call from a car or somewhere noisy.
7. Say “call me back later” without giving a time, then check that a callback actually got scheduled. Our own agent failed this on a live call in October: it promised to call back later that day, and nothing put the call on the schedule. We’ve fixed it since.

Measure it against voicemail

I wouldn’t trade a great receptionist who knows every customer by name for software, mine included. But if your calls go to voicemail while the team does paid work, you’re weighing under $100 a month at low volume against the jobs that go to whoever answered first.

Once it’s live, read the transcripts for the first few weeks and add answers to the questions you never covered.