AI receptionists for small business: how they work, what they cost, and when they make sense.

What an AI receptionist actually is

An AI receptionist is a voice agent that answers your business phone line, understands spoken requests, and responds in a synthesized voice, without a human involved unless the call needs one. Mechanically, it's three things stitched together: speech-to-text to turn the caller's words into text, a language model to decide what to say or do, and text-to-speech to turn the response back into audio, all over a live connection to your phone line.

OpenAI's Realtime API, for instance, is built specifically to "listen, reason, speak, and call tools" over an open low-latency connection, and its documentation explicitly recommends it for telephony voice agents using a SIP connection, the same protocol that routes traditional phone calls over the internet. Twilio's ConversationRelay does the equivalent job for developers building on Twilio's phone infrastructure: it connects an inbound call to your AI application over a WebSocket and handles the speech-to-text and text-to-speech conversion in between. Both platforms support interruption handling, meaning a caller can talk over the AI mid-sentence the way they would with a person, rather than waiting for a menu to finish.

For a small business, this means the AI can do more than a phone tree ever could: answer "what are your hours," check an appointment slot, take a message with details, or transfer to a specific person, all through normal conversation instead of navigating a menu.

AI receptionist vs answering service vs human receptionist

There are three staffing models here, and they land at different price points for different reasons.

A human receptionist, employed directly, costs a median of $37,230 a year (about $17.90 an hour) according to the Bureau of Labor Statistics, on top of whatever benefits and overhead your business carries. That figure covers roughly a million receptionist jobs nationally, and BLS projects essentially flat employment growth in the role over the next decade, which tells you it isn't a shrinking function so much as a stable, unglamorous one.

A live answering service outsources that person but keeps a human on the line. Smith.ai charges $300 a month for 30 calls up to $2,100 a month for 300 calls, with per-call overage between $8.50 and $11.50 depending on the tier. Ruby prices by receptionist minutes instead of calls: $250 a month for 50 minutes up to $1,725 a month for 500 minutes, both with 24/7 coverage and no setup fees. These services can use AI-first or human-first handling at the same rates, which is itself telling: the pricing floor is set by staffing a person to be available, not by the technology.

An AI receptionist strips out the live human and prices by usage instead, which is why it undercuts both of the above at volume, covered in the next section. The honest comparison isn't that AI is cheaper across the board; it's that AI is cheaper per call once you clear a certain volume, and a live human is better at judgment calls and unusual situations no matter the volume.

What it costs: platforms, per-minute pricing, and hidden fees

AI phone answering pricing splits into three models. Vertical, ready-to-use products charge per agent or per location: Goodcall runs $79 to $249 a month per agent with unlimited minutes and tokens but caps monthly unique customers at 100, 250, or 500 before charging $0.50 per additional customer (annual billing brings those tiers to $66, $108, and $208). Slang.ai, built specifically for restaurants and integrated with OpenTable, SevenRooms, and Yelp, charges $399 a month per location for its Core plan and $599 for Premium, which adds bilingual Spanish support and missed-call capture.

Developer platforms charge per minute instead, which is cheaper at low volume but requires someone to actually build the agent. Retell AI's all-in cost runs $0.07 to $0.31 per minute depending on the model, made up of voice infrastructure, text-to-speech, telephony, and the language model itself. Vapi charges a flat $0.05 per minute platform fee and passes through the language model and speech costs at cost, which can be $0 if you supply your own API key; it includes 10 concurrent call lines before charging $10 per additional line per month.

Then there's the enterprise tier: Synthflow AI has dropped small-business pricing entirely and now lists only a custom Enterprise plan starting at $30,000 a year, scoped to call volume and integrations. That's a useful signal on its own; some AI-voice vendors have decided small businesses aren't their market anymore, which narrows the realistic small-business field to the per-minute developer platforms and the vertical per-agent products. Our breakdown of how much AI automation costs walks through the same build-vs-buy tradeoff for other automation projects, not just phone answering.

What a missed call actually costs you

The case for answering every call, by whatever means, rests on how expensive a missed one is. Platform data from Invoca puts the overall unanswered-call rate at about 26%, and over 60% in some industries. Qualified phone leads convert at an average rate of 41%, which is why an unanswered ring is a real leak, not a minor inconvenience. Voicemail doesn't rescue much of that: fewer than 3% of callers who reach voicemail actually leave a message, and the rest just hang up.

Invoca's own case data on an automotive repair chain with more than 1,200 locations found it was missing over 8,000 calls a month, which the company estimated at a $16 million annual leak at a $220 average ticket. That's one company's numbers, not a universal multiplier, but it illustrates the mechanism: at scale, a percentage of missed calls compounds into a large number fast, even at a modest average ticket size.

Where AI receptionists break down

No vendor in this space claims full autonomy, and it's worth taking them at their word. Slang.ai describes its own restaurant product as one that "instantly answers common questions" while routing complex or sensitive inquiries to staff. That's the honest boundary: an AI receptionist is good at repetitive, well-defined requests, hours, availability, simple bookings, basic FAQs, and it's meant to hand off anything ambiguous, emotional, or high-stakes.

The interruption handling in platforms like Twilio's ConversationRelay (controlled through settings like "interruptible" and "interrupt sensitivity") makes conversations feel natural, but naturalness isn't the same as judgment. A caller who's upset or asking something outside the script needs a person, and the systems available today are built around that handoff, not around eliminating it. If your call volume is mostly complex or relationship-driven, like a law firm fielding new-client intake or a home-services company juggling emergency dispatch, an AI receptionist will frustrate more callers than it helps. If it's mostly scheduling and simple questions, it fits. Our guide to AI automation for small business covers this same pattern-matching exercise for other automation candidates: automate the repeatable part, keep a human on the judgment calls.

Compliance basics: consent and call recording

Two separate rules matter here, and businesses conflate them often enough that it's worth separating clearly.

First, recording. Federal wiretap law allows recording a call if you're a party to it or if one party has consented, known as one-party consent. Some states require all-party consent instead, so check your specific state's statute before you turn on recording by default in an AI receptionist platform; there's no single authoritative government list, so this is a case where you need to verify locally rather than trust a generic answer.

Second, outbound calls. If your AI receptionist places calls out, like reminders or confirmations, FCC rules require any prerecorded or AI-generated voice message to identify the caller's name, number, and business name at the start of the message. The FCC has also stated plainly that AI-generated voice calls are illegal unless the consumer has agreed to receive them or the caller qualifies for an exemption, and consumers can revoke consent at any time and in any reasonable manner. Answering inbound calls doesn't trigger the same outbound-consent requirement, but if you're layering in outbound reminder calls, get written or clearly documented consent first.

Do reminders and follow-ups actually work

One of the more provable benefits of an AI receptionist, or any automated calling and texting system, is reducing no-shows through reminders. A systematic review and meta-analysis across ten randomized controlled trials found patients receiving appointment reminders were about 11% more likely to attend than those who received none. A separate randomized trial at a pediatric clinic found text message reminders cut the no-show rate from 38.1% in the control group to 23.5%, a meaningful gap on a modest sample.

These numbers come from healthcare studies, not a specific AI receptionist deployment, so treat them as evidence that reminders work as a mechanism, not a promise about what any particular vendor's product will do for your business. If your business loses revenue to no-shows, an automated reminder workflow, whether it's voice, text, or both, has real research behind it. That's a more defensible reason to automate than the idea that AI sounds impressive.

How to decide if you need one, and what to do first

Start with your call pattern, not the technology. Count how many calls you miss in a typical week, how many are routine versus complex, and what you're currently spending on any answering solution, live or automated. If missed calls are common and most of what callers want is simple, checking hours or booking a basic appointment, an AI receptionist is worth piloting, especially given how cheaply the per-minute developer platforms scale compared to a live answering service's per-call rates. If your calls skew complex or emotionally sensitive, put your money into a live answering service or in-house staff instead, and consider AI only for after-hours overflow.

Before you sign anything, get clear on the consent and recording rules for your state and your outbound call plans, since retrofitting compliance after deployment is harder than building it in from the start. What happens after the call matters just as much: routing messages, updating a CRM, or triggering a follow-up is its own project, covered under Workflow Automation. Async Automations starts every engagement, including AI receptionist projects, with a free audit of current call handling and workflow.