AI receptionist for small business: what it really costs.

An AI receptionist for small business owners answers the phone when you cannot: it greets the caller, answers plain questions, takes a message, books an appointment or puts an urgent call through to you. A plumber is under a sink, a dentist is with a patient, a solicitor is in a meeting, and the phone rings out. Most owners know that costs them something and few have put a number on it, because the usual fixes look expensive: hire a receptionist, pay a live answering service or trust a robot that might embarrass you.

The published prices are strikingly low, from $0 for a trial to a few hundred dollars a month. Cheap does not mean simple. Vendors count usage four different ways, and the same 100 calls can cost anything from $49 to about $900 depending on the unit your bill is built on and whether a human or a model answers. We set up AI receptionists for a living, so this guide gives you the prices straight from each vendor's own page, read on 17 October 2026, turned into a monthly bill at 20, 100 and 500 calls, plus what a price list never tells you: where these systems fail, what the law says and what to ask before you sign. If you would rather have one set up and tested for you, our AI receptionist automation service is fixed price and paid once.

What is an AI receptionist, and what are you actually buying?

A receptionist does three jobs, and it helps to separate them because AI products are good at some and poor at others.

1. Pick up and say something sensible. Greet the caller, work out what they want and answer plain questions: opening hours, the address, what you do, whether you cover their area.
2. Do something useful with the call. Take a message, book into a diary, send a link by text, or transfer the call to a person.
3. Leave a record. Write down who called, what they wanted and what was agreed, in a place your team will actually look.

A good AI receptionist does the first reliably, the second when it is connected to your diary and phone system, and the third only if somebody has wired it into the tools you use. Those connections are the difference between a service that answers calls and one that improves your business, and they are where most of the cost and most of the failures sit. It is why our AI call transcription and CRM logging service exists: it writes each call's summary, outcome and next step onto the right contact record.

There are three broad ways to get one. A hosted plan is a product you subscribe to and configure with a prompt and a few settings. A human answering service is people, increasingly with AI on the side. A built agent is a voice agent you assemble on a developer platform, which gives full control and a bill made of several components. The prices and risks differ, so we take them in turn.

AI receptionist pricing: how vendors count usage

Before any price, the unit. It is the single biggest reason two quotes for the same business look unrelated.

Two consequences follow. A product that bills by the call is insulated from long, rambling conversations and exposed to a surge of short ones. A product that bills by the minute is the reverse. If your callers mostly ask one quick question, minutes are cheap. If they tell you their life story, they are not.

AI receptionist cost: hosted plans compared

These four vendors publish a price list and are representative of how the hosted category prices. We read each page on 17 October 2026. The figures are list prices in US dollars on monthly billing.

Smith.ai's page also lists Enterprise plans at higher volumes with custom prices, and Goodcall lists an Enterprise tier without a price. Rosie's page shows a Growth plan we do not quote because we could not confirm its minutes. Dialzara sells outbound voice, SMS and website chat as separately priced add-ons, and Rosie's website texting add-on is $50 a month on top of its plan, so an add-on is where a $49 plan can become a $99 one.

Human answering service cost compared

Human services are priced for people, so they cost an order of magnitude more per call. They are also the benchmark the AI plans are measured against.

Ruby's page states that optional AI enhancements are available on all plans at no extra cost, and that there are no extra fees for activation, onboarding, setup or customisation. Abby's page offers a Human plus AI receptionist that lets you choose how each call is handled. The line between human and AI services is blurring: several vendors now sell both and let you route a call to either.

What an AI receptionist costs at 20, 100 and 500 calls a month

To compare them we need one set of assumptions. We assume the average call lasts 2 minutes of talk time. That is our assumption, not a measured figure, and it matters for every per-minute plan, so a sensitivity table follows. For each vendor we pick the cheapest published plan that covers the volume and add the published overage rate where the plan runs out.

"Not published" means the volume exceeds the largest plan on the vendor's page and the page gives no overage rate, so we cannot compute a bill without inventing one. The last row is Retell's own worked example of $0.11 a minute multiplied by the minutes. It is the cost of the calls only, with no build, phone number or anybody's time.

What if your calls run longer than two minutes?

Per-minute pricing is the only kind that punishes long calls, so it deserves its own sensitivity table. Here is Dialzara at 100 calls a month for three average call lengths, using its published plans and overage rates.

Doubling call length from two minutes to four takes the bill up by about 87 per cent, from $96.20 to $180. Smith.ai's per-call price does not change at all. You cannot know call length in advance, so if your callers talk at length, price on calls or customers, not minutes.

Build your own AI receptionist on Vapi, Retell or Bland

The third route is to assemble a voice agent on a developer platform. It has the lowest running cost and the highest set-up cost, and it is the right answer when the agent has to do something a hosted plan will not: read from your own database, follow your own booking rules or sit behind your own phone system.

Platform pricing is built from parts. We read three platforms' own pages and Twilio's telephony page.

For comparison, Twilio's US voice page, marked current as of August 2026, lists inbound calls to a local number at $0.0085 a minute plus $1.15 a month for the number. Telephony is the cheapest component and the one people forget to add.

Multiply by our 2-minute assumption and these are small numbers: roughly $22 a month at Retell's $0.11 example for 100 calls, about $110 for 500. The running cost does not decide this route. The build does. Prompts, call flows, booking logic, connection to your diary or CRM, testing against real calls and upkeep are all human work, priced separately. In our own catalogue a platform setup is a one-off service: Vapi voice agent setup is $2,875 to $6,725, Retell voice agent setup is $1,940 to $4,535, Bland phone agent setup is $2,130 to $4,980, and a platform-neutral AI voice agent setup is $1,750 to $4,090, each paid once.

AI receptionist vs hiring a receptionist

The US Bureau of Labor Statistics gives a median pay for receptionists of $38,010 a year and a median hourly wage of $18.27 in May 2025. Those are wages only, and the page we read does not include payroll tax, benefits, equipment or cover when the person is ill or on holiday, so the true cost is higher.

The figure that gets overlooked is coverage. A week has 168 hours. A person working 40 of them covers about 24 per cent of the week, and the other 76 per cent is evenings, weekends, lunch breaks and sickness. An AI plan or an answering service answers at 3 a.m. on a Sunday. A single employee does not.

The conclusion is not that a receptionist is poor value. For a business with walk-in customers and a lot of in-person work, an employee does things no phone service can. The point is that the two are rarely substitutes: the phone service covers the hours the employee is not there, and the employee covers the work the phone cannot do.

What a missed call is worth to your business

A price only means something against value. Nobody can tell you what a missed call costs your business, because it depends on your price, your close rate and how many callers would try another number. There is one published statistic worth knowing, with limits. Moneypenny, a UK answering firm, published a small-business call report in November 2016, based on an independent survey of 300 micro businesses and data on 10,000 businesses' calls. It found that 69 per cent of callers who reached voicemail hung up without leaving a message. It is a vendor survey, from 2016, in the UK, so treat it as a reason to check your own numbers, not as a benchmark.

Here is a way to do the sum with inputs you can find in your phone records and your books.

With these illustrative inputs a hosted plan at about $100 a month looks like good value and a $2,100 human plan does not. With a first-job value of $3,000 instead of $300, both look cheap. The tool is not the variable. Your job value is. A plan that costs $99 and recovers one job a quarter has paid for itself, and one that recovers none has not, however good the demo sounded.

If the pain is the missed call itself rather than the conversation, a lighter fix may be enough: an automatic text to the caller that says you will ring back and asks what they need. It costs far less than an AI conversation, and we sell it as a service for several trades, for example missed-call text-back for home services and for salons.

Should an AI answer every call, or only the ones you miss?

Most small businesses do not need an AI to answer every call. A cheaper and safer pattern is to let people answer when they can and send everything else to the AI: after hours, at lunch, when the line is engaged or after a few rings. That changes the maths in three ways.

• Volume falls. If you answer half your calls yourself, the plan you need may be one tier down, and per-call and per-customer plans fall in direct proportion.
• Risk falls. The calls the AI handles are the ones that would have gone to voicemail anyway, so a mistake is rarely worse than the status quo.
• Trust is easier to build. You hear what the agent does with the overflow before you trust it with the whole front door.

The trade-off is that a caller who reaches a person on one call and an AI on the next may notice the difference. A greeting that says what the AI is, and a promise that someone will follow up, goes a long way. Human services work the same way. Their plans are small allowances, 30 to 300 calls a month at Smith.ai, so they suit overflow and after-hours better than every call. If you route every call through a live service at $8.50 to $11.50 for each call beyond the plan, the bill climbs quickly, as the 500-call row shows.

Where AI receptionists go wrong: seven calls to test

The price list is the easy part. The harder question is what happens on the calls that are not routine. These are the failure points we test before trusting an AI with a front door. They are our assessment of the category, not measured failure rates for any vendor.

A sensible rollout runs the AI alongside a person for a while, reads the call summaries every day at first and has a fast way to correct the agent when it gets something wrong. Our automation brief guide is a good template for writing down what the agent should and should not say.

AI receptionist legal basics: consent, recording and privacy

This section is general information, not legal advice. Laws differ by country and state, and a few minutes with a qualified adviser is cheap compared with getting it wrong.

AI voices and the US telephone law

In February 2024 the US Federal Communications Commission adopted a Declaratory Ruling that calls made with AI-generated voices are "artificial" under the Telephone Consumer Protection Act. The ruling took effect immediately. In practice the rules about artificial and prerecorded voice calls, including the need for the called party's prior express consent for many kinds of call, apply to AI-voiced calls. It matters most for outbound calls, such as an AI agent calling back a lead or chasing a payment. An AI answering your inbound calls is a different situation, but if you also use it to call out, take advice on consent.

Recording and transcribing calls

Most AI receptionists record calls and transcribe them. In the United States the rules on recording differ by state, and some require the consent of everyone on the call, not just one party. We have not listed which states, because lists differ between sources and change, and the safe course is to follow the strictest rule that could apply to the caller and tell every caller at the start. A short spoken notice, and a vendor setting that plays it, is cheap insurance.

Personal and health information

If callers give health details, financial details or identity numbers, ask the vendor where the data is stored, how long it is kept, who can hear the recordings and whether they will sign a data processing agreement. For US health practices, ask in writing whether the vendor will sign a business associate agreement before any patient detail passes through the system. We could not read the regulator's own guidance page when preparing this guide, so confirm the requirement with your adviser rather than relying on our summary.

Telling callers it is an AI

Some jurisdictions already require or are considering rules about disclosing an AI voice, and callers generally feel better when told. We recommend a plain greeting that says what it is. It costs nothing, protects you and sets the right expectation about what the agent can do.

Five AI receptionist buying mistakes

1. Comparing monthly prices without comparing units. A $79 plan counted by customer and a $99 plan counted by minute are different products, and the cheaper-looking one may be dearer at your volume.
2. Skipping the overage line. The plan that is cheapest in a quiet month can be the dearest in a busy one.
3. Judging by the demo. A scripted demo call is the easiest call there is. Ask to hear recordings of ordinary customers, or run a trial on your own number.
4. Forgetting the connection. An AI that takes a message and emails it is a smarter voicemail. The value arrives when it books into your diary or logs the call in your CRM, and that connection is usually an extra cost.
5. Assuming it is finished once it is live. Menus change, prices change and staff leave. Someone must own the prompt, read the summaries and fix mistakes, and that person's time is part of the cost.

What to measure in the first month

If the cost per captured booking is below the profit on one booking, the receptionist is paying for itself. If most calls end in a transfer, the agent is a router, and a cheaper phone tree or a text-back may do the same thing for less.

Questions to ask any AI receptionist vendor

Which AI receptionist route fits your business?

If you sit in the fourth or fifth row, an industry-specific build saves a lot of explaining. Our dental AI receptionist, medical practice AI receptionist and law firm AI receptionist services come with the sorting, routing and escalation rules for those trades already thought through, and the cost model is the one in this guide. To see what an automation of this kind is worth before you commit, our guide to how much business automation costs gives the payback model.