AI receptionists for small businesses: what they cost, what they handle and how to set one up

Picture a two-bay auto repair shop. The owner is under a car with both hands busy when the phone rings for the fourth time this hour. One caller wants to know if the shop does brake jobs, one wants to move an appointment, one is a supplier and one has a car that just died on the highway. He can answer none of them. By the time he calls back, at least one has booked somewhere else.

That is the problem AI receptionists solve. They answer every call in a natural-sounding voice, any hour of the day, handle the routine questions, book appointments, take messages and send the urgent ones to a person. Today it can be a monthly plan you set up in an afternoon, with free tiers to test on real calls.

What an AI receptionist actually does

Most services follow the same pattern. You give the AI your business details, services, hours, prices you are happy to share and rules for what to do with different kinds of calls. It answers in your business's name, works out what the caller needs, answers from what you gave it, and then books, takes a message, transfers the call or texts you a summary. Smith.ai, for example, lists 24/7 answering, recording with transcripts and summaries, lead qualification and call routing, scheduling through tools like Calendly, and a test studio so you can try it before going live. ¹

The difference from an old-style phone menu is that callers just talk. Nobody presses 2 for service. Done well, most callers either get their answer or get to the right person faster than before. Done badly, it traps people in a loop with a confident voice that cannot help them, which is worse than voicemail.

What to hand it, and what to keep

The single most important decision is which calls the AI finishes and which it hands to a person. A simple rule: let it handle calls where the right answer is written down somewhere, and route calls that need judgment, money or empathy.

What it costs

Pricing usually works one of two ways: a monthly plan with a set number of calls, or credits you spend per call. Here are two real examples at different ends of the market. Smith.ai's AI receptionist starts free with 25 calls a month, then $3 per extra call. Its Pro plan is $150 a month for 75 calls, about $2 a call, and its Enterprise plan, with a done-for-you setup, works out to about $1.67 a call at 300 calls a month. ¹

Quo, a business phone service, sells its Sona AI agent in credit tiers on top of a Quo phone plan. Each workspace gets about 10 calls a month free; paid tiers run from $25 a month for about 40 calls to $199 a month for about 600. Calls under 15 seconds do not count, and by default Sona stops answering when credits run out rather than billing you extra, sending calls to your fallback instead. ²

Those two are not like-for-like. Smith.ai is a standalone answering service, with dedicated onboarding on its paid plans. Sona is cheaper per call but lives inside Quo's phone system, so it suits businesses already using, or willing to switch to, Quo for their phone lines.

Whether it pays depends on what a missed call is worth to you. Here is an illustration, not a measurement: if the repair shop misses 10 calls a week and even two of those were jobs worth $250 each, two jobs a week are walking away. Against a plan of $50 to $150 a month, the AI only has to save one job a month to cover itself. Check your own phone logs for missed calls before you buy.

The legal basics: recording and AI voices

Most AI receptionists record and transcribe calls, which raises consent. Washington, where Monolith is based, requires the consent of everyone on a call before it is recorded. The law treats consent as given when one party announces, in a reasonably effective way, that the call is about to be recorded, and that announcement must itself be recorded. In practice, that means your AI's greeting should say the call is recorded. Some other states have similar all-party rules, so check yours. ⁴

Outbound calls are stricter. In 2024 the FCC ruled that calls using AI-generated voices count as "artificial" under the Telephone Consumer Protection Act, the federal law that already restricts robocalls. If your system will make outbound calls, such as appointment reminders or callbacks, make sure you have the consent the law requires before switching that on. This is a summary, not legal advice; check with your attorney for anything beyond answering inbound calls. ³

Set one up in a week

The businesses that get good results treat setup like training a new front-desk hire: clear instructions, a test period and a weekly check. Each step below has a check before moving on.

Mistakes to avoid

• Pretending to be human. Let the greeting say it is an AI assistant; callers forgive a robot that helps, not one that deceives.
• No way out. Callers who ask for a person should reach one, or at least a fast callback.
• Inventing answers. Limit it to your written information, and tell it to take a message when it does not know.
• Forgetting the recording notice. Put it in the first sentence of the greeting.
• Never checking. Read the summaries weekly for wrong answers, missed bookings and patterns worth fixing.