Most writing about AI receptionists is aimed at companies with a procurement process. If you are three people and one of you is currently holding a phone with wet hands, almost none of it applies.
Here is the version for a business where the owner answers the phone.
What is different about a small business
There is no IT person. If setup takes a project plan, it will not happen. The realistic budget for getting this working is an afternoon, and most of that afternoon is you on a call with someone showing you where things are.
The phone competes with the work. In a larger business, answering the phone is somebody's job. In yours it is an interruption to the thing that earns money. That is the actual problem being solved — not call volume, but the fact that every ring is a choice between the customer in front of you and the one on the line.
A bad month hurts. Usage-based pricing that triples in December is a genuine cash problem rather than a line item, which is why the pricing model matters more to you than it does to a chain.
You cannot afford a bad answer. A large company with a bad automated system annoys people. A small one loses a regular, and the regular tells people.
The three things worth paying for
It completes the booking. Not a message, not a callback list. If the system cannot see your live calendar and write into it, you have bought a more expensive voicemail. This single question eliminates about half the market.
It knows your real menu. Your services, your durations, your prices, who can do what. Not a script somebody typed at onboarding and never updated — a system quoting a price you stopped charging six months ago will do it in front of a customer.
It admits what it does not know. The most underrated feature. A system that invents an answer about a price, an allergy or a medical question is more dangerous than one that says a person will call back, because it sounds confident.
The three things not worth paying for
Voice cloning. Making it sound like you specifically. Callers work out it is a machine within a sentence regardless, and most of them do not mind — what they notice is that somebody picked up at eight on a Sunday.
A hundred integrations. You need one: the calendar you actually use. The rest is a feature list.
Outbound calling. Sold as a bonus. In the United States automated outbound calling to consumers carries real telemarketing obligations around consent, calling hours and do-not-call lists, with penalties per call. Unless you have taken advice, this is a feature to leave switched off.
What it should cost
Model it against your own numbers rather than a vendor's example.
Usage-based pricing exposes you to your own busy season. Flat pricing does not. For a business with seasonal peaks — and almost every appointment business has them — that difference is worth more than a few dollars of headline rate. We priced ours flat by team size for exactly that reason, from ninety-nine dollars a month for up to five staff, and the pricing page has the tiers.
The comparison that decides it is not against another vendor. It is against what the missed calls are already costing. Count the calls you miss in a week, take the share that wanted an appointment, multiply by your average ticket. Most owners find that number is larger than any of the prices they were comparing.
Start with the overflow, not the whole line
The mistake is treating this as a decision about your phone. It is not, and framing it that way makes it feel far riskier than it is.
Conditional call forwarding sends only the calls you do not answer — the ones already reaching voicemail — wherever you choose. Your number does not change. Your existing setup keeps working. Nothing that currently works stops working. The codes differ by carrier and they are in the call forwarding guide.
So the smallest useful version is: point the calls that are currently reaching a beep at something that can book, and change nothing else. Two weeks of that tells you more than any demo, because it is measured on your real callers.
The test that matters
Before you commit, ring whatever you are considering and ask it for something it cannot know — a price for something not on the menu, or whether a treatment is safe in some specific circumstance.
If it admits it does not know and offers a callback, it is safe to put in front of your customers. If it improvises, it will eventually improvise at someone who believes it. There are four more tests like this in five calls to make before you trust an AI receptionist, all of them failures we found in our own system first.