What a small business AI receptionist actually does
Answers every inbound call with a natural-sounding voice, books or reschedules appointments, screens and routes leads, and transfers anything it can't handle to a real person.
A small business AI receptionist sits on your main phone line through a forwarded or ported number, answers in your business's voice and tone, and runs a loop on every call: transcribe the caller's speech, understand the intent (booking, question, complaint, sales pitch), check your calendar or knowledge base through a tool call, either complete the task or gather details, and either hang up with a resolution or transfer to a human with a summary of the call. Behind the packaged product sits a voice stack: speech-to-text, a reasoning model, text-to-speech, and a telephony layer that actually carries the call, usually billed separately when you build it yourself and bundled into one flat rate when you buy.
This market has moved fast through 2026. Goodcall's official pricing is per unique caller rather than per minute, a structural bet that most small businesses care more about not missing a caller than about call length. Meanwhile the developer-platform layer (Vapi, Retell, ElevenLabs Agents) kept per-minute rates roughly flat to slightly down even as voice quality and latency improved, which is why a growing number of agencies now build custom receptionists instead of reselling a packaged SaaS tier.
Where it still breaks: long, detailed quotes that require real judgment, a genuinely upset caller who needs empathy instead of a script, and anything that depends on reading context a human would pick up instantly (a regular customer's tone, a half-finished sentence). The often-cited "62% of small business calls go unanswered" statistic traces back to a 2016 study of 85 businesses by 411 Locals, an old and small sample, but it's the number this entire product category was built to fix, and it's still the most-cited figure in the space.
AI capabilities involved
Real-time speech-to-text on the live call
Understanding intent and holding the conversation
Natural-sounding voice output (text-to-speech)
Booking, rescheduling and calendar lookups via tool calls
Who uses this
- Single-location service businesses (salons, auto shops, cleaning services) answering 50-500 calls a month with no dedicated front-desk staff
- Solo and small-team professionals (contractors, consultants, local law or accounting practices) who lose business to voicemail during job sites or client meetings
- Multi-location franchises standardizing after-hours and overflow coverage across every site
- Businesses already paying for a human answering service and shopping for a cheaper 24/7 alternative
What the small business AI receptionist handles (and what it shouldn't)
| Task | Agent handles it? | Systems it touches | Human role |
|---|---|---|---|
| Answering every call so nothing goes to voicemail | Fully | Phone number, telephony carrier | None |
| Booking and rescheduling appointments | Fully | Calendar (Google/Outlook/Calendly) | None for standard slots |
| Screening and qualifying sales or lead calls | Fully | CRM | Reviews flagged high-value leads |
| Answering hours, pricing and service FAQs | Fully | Knowledge base or scripted answers | None |
| Taking a message and routing an urgent call | Fully | Call transfer, SMS/Slack alert | Picks up the transfer or reads the message |
| Quoting complex or custom pricing | No | None | Takes the call or calls the lead back |
| Handling an angry or distressed caller | No | Call transfer | Takes over immediately |
| Collecting a card number for a deposit | With approval | Payment link via SMS, not the call itself | Confirms payment landed |
Off-the-shelf AI agents on the market
Real products you can sign up for today — with current 2026 pricing, honest pros and cons.
Goodcall
Single-location businesses that want predictable pricing regardless of call length
No free tier; plans start at Starter
Starter $79/mo (100 unique callers, unlimited minutes); Growth $129/mo (500 callers); Scale $299/mo (2,000 callers)
Custom enterprise tier for higher caller volume
Pros
- +Billed per unique caller with unlimited minutes, so long calls don't spike your bill
- +Google Voice and Zapier compatible out of the box
- +Overage drops to as low as $0.15/extra caller on the Scale plan
Cons
- −External tool integrations are capped by tier (1, 5 or 15 tools)
- −No published HIPAA or legal-vertical tier
Smith.ai AI Receptionist
Law firms and professional-services businesses already on Clio
Free plan: 25 calls/mo, $3/call overage
Pro $150/mo (75-300 calls, about $2/call); Enterprise $500/mo (300-1,000+ calls, about $1.67/call)
Enterprise tier scales past 1,000 calls/mo
Pros
- +Native Clio integration makes it the strongest option for small law firms
- +Month-to-month with 30-day notice, no annual lock-in
- +Native Zapier and Calendly connections
Cons
- −Per-call pricing punishes high call volume compared to Goodcall's per-caller model
- −The free tier's $3/call overage adds up fast past 25 calls
Rosie
Very low-volume businesses just getting started with an AI receptionist
No free tier
Professional $49/mo (250 minutes); Scale $149/mo (1,000 minutes); Growth $299/mo (2,000 minutes)
Not published; contact sales for higher volume
Pros
- +Cheapest entry tier among packaged receptionists
- +Booking and call transfers available from the Scale tier up
- +Annual billing gets two months free
Cons
- −Overage pricing past the included minutes isn't published
- −Booking isn't included on the cheapest Professional tier
My AI Front Desk (Frontdesk)
Businesses serving a multilingual customer base on a flat monthly budget
No free tier
Business-in-a-Box $99/mo ($79/mo billed annually), includes 200 voice minutes
Partner/Enterprise tier adds API access
Pros
- +Supports 20+ languages out of the box
- +Flat pricing is simple to budget against
- +Zapier integration for routing bookings and messages
Cons
- −Overage runs about $0.25/minute once you exceed the included 200 minutes
- −API access is Partner/Enterprise only, limiting custom integration
The AI stack
A voice receptionist needs a real-time voice layer (STT and TTS), a reasoning model, telephony to actually carry the call, and the same observability layer any text agent needs.
Voice (speech-to-text + text-to-speech)
Converts the caller's speech to text for the model and converts the model's reply back to natural speech
ElevenLabs Agents
$0.08/min on every self-serve plan; $0.16/min burst over your concurrency capVoice quality as the top priority
Deepgram + OpenAI Realtime
Deepgram STT plus gpt-realtime-2.1 at roughly $0.04-$0.10/min combinedDeveloper-assembled stacks wanting best-in-class transcription
Bland AI (bundled)
$0.11-$0.14/min all-in, STT/LLM/TTS included in one rateTeams that want one vendor instead of three
Our pick: ElevenLabs Agents for voice quality if you're assembling your own stack; a bundled vendor like Bland if you'd rather have one invoice than three.
Telephony
Carries the actual phone call and connects it to your voice AI pipeline
Vapi's built-in telephony
Free SIP/number hosting; Twilio pass-through at $0.008/min inboundDeveloper builds on Vapi
Twilio direct
$0.008/min inbound, $0.014/min outboundFully custom builds
Our pick: Let whichever orchestration platform you pick (Vapi, Retell) handle telephony pass-through rather than managing Twilio directly, unless you already run Twilio elsewhere.
Reasoning model
Understands the caller's intent, holds the conversation and decides when to book, transfer or end the call
Claude Sonnet 5.5
$2/M input, $10/M outputThe main conversation loop on a custom build
GPT-6.1 Sol via gpt-realtime-2.1
Realtime audio pricing: $32/M audio-in, $64/M audio-out tokens, roughly $0.04-$0.10/min in practiceLowest-latency custom voice builds
Our pick: A chained Sonnet 5.5 + ElevenLabs stack for predictable tool-calling, or OpenAI's realtime speech-to-speech model when latency matters more than architectural control.
Observability & evals
Reviews call transcripts for missed bookings, bad transfers and compliance-script drift
Vendor call logs and transcripts
Included on most SaaS tiersBuy-SaaS
Langfuse
Free up to 50K units/mo, then $29+/moCustom builds
Our pick: Review a random sample of real call recordings weekly no matter which path you pick; voice agents drift in ways text transcripts alone don't always show.
Reference architecture
The hardest engineering problem is latency: every extra 300-500ms between the caller finishing a sentence and the agent responding reads as an awkward pause on the phone, so the STT, model and TTS legs all have to be tuned together rather than optimized in isolation.
Call connects
Telephony (Twilio/Telnyx or vendor-native)The caller's phone call is routed to the AI receptionist's number
Speech-to-text
STT engine (Deepgram, ElevenLabs Scribe)The caller's audio is transcribed to text in real time, streaming as they speak
Intent and reasoning
Reasoning modelThe model reads the transcript, decides the intent, and calls a tool if it needs calendar or account data
Tool call
Calendar/CRM integrationA booking, availability check or lead-qualification lookup runs against your real system
Text-to-speech
TTS engine (ElevenLabs, Deepgram Aura)The model's reply is converted back to natural-sounding speech and streamed to the caller
Transfer or escalation
Call-transfer logicAngry callers, complex quotes or anything outside the script get warm-transferred to a human line
Call logging and eval
Observability toolThe recording, transcript and outcome are logged for a weekly quality review
Estimated cost per request
~$0.08-$0.15 per minute of call on a self-assembled stack (Vapi, Retell or ElevenLabs plus LLM and telephony); roughly $1.67-$2.00 per call on packaged SMB SaaS tiers
Cost calculator
Drag the sliders to model your actual usage. The numbers update in real time so you can stress-test economics before writing a single line of code.
Estimate your monthly cost at your own call volume and average call length.
Estimated monthly cost
$239
≈ $2,868 per year
Calculator notes
- The packaged SaaS line uses Goodcall's Starter overage rate ($0.79/extra caller) as a conservative per-call proxy; most bundled plans are cheaper per call at low volume because the base fee covers the first 100+ callers.
- The DIY line assumes a 3-minute average call on an all-in Vapi/ElevenLabs/Deepgram stack at roughly $0.11/min; adjust the average-call-length slider if your calls run longer.
- A custom agency build replaces both lines with your own voice-stack bill, typically $50-$300/mo at this volume, plus the upfront build cost.
- None of these figures include a human receptionist's time; Ruby's human-staffed service runs roughly $3.45-$5/min for comparison.
Build it yourself with vibe-coding tools
Vapi lets you assemble a working AI receptionist from a phone number, a system prompt and a calendar tool without writing a full backend.
Time to MVP
A few days for a basic version, 2-4 weeks to get call quality, transfers and consent language production-ready
Total cost to MVP
$0-$100 (a Vapi Core plan at $29/mo plus usage, or pay-as-you-go with no monthly fee)
You'll need
Starter prompt
Create a new Assistant in Vapi using Claude Sonnet 5.5 as the model, ElevenLabs for text-to-speech, and Deepgram for transcription. Set the first message to: 'Thanks for calling {Business Name}, this is their AI assistant, how can I help you today?' so the disclosure happens before anything else is said. Give the assistant this system prompt: 'You are the phone receptionist for {Business Name}, a {business type} in {city}. Greet callers, answer questions about hours, pricing and services using only the information provided below, and help callers book, reschedule or cancel appointments using the check_availability and book_appointment tools. If a caller asks for a detailed custom quote, sounds upset or asks for something outside what you can do, say: "Let me connect you with someone who can help with that right away," and use the transfer_call tool to send them to {human phone number or line}. Never claim to be human if asked directly. Keep responses short and conversational, the way a real receptionist would talk on the phone, not like a written document. At the start of the call, this line is recorded for quality purposes, consistent with the greeting already given.' Add two tools: check_availability and book_appointment, both pointed at your calendar's API (or a Zapier webhook if you don't want to write direct API code), and a transfer_call tool pointed at a mobile number or your existing office line. Set call recording on, and configure the transfer to carry a short AI-generated summary of the call to whoever picks it up.
Paste this into Vapi
Follow-up prompts (run in order)
- 1
Add a tool that looks up an existing customer's appointment history before booking a new one
- 2
Add a Spanish-language path that detects the caller's language and switches voices
- 3
Add an SMS follow-up that texts the caller a confirmation after every booked appointment
- 4
Add a nightly report that lists every call the assistant transferred and why
- 5
Add DTMF keypad capture so a caller can pay a deposit without speaking a card number aloud
Expected output
A phone number that answers every call within one ring, books real appointments onto your calendar, and warm-transfers anything it can't handle with a spoken summary for the person picking up.
Known gotchas
- !Latency stacks up fast: STT plus model plus TTS plus telephony can easily cross 1 second round-trip if you don't choose fast providers at each layer, and callers notice anything over about half a second of dead air
- !Voice providers' advertised latency (Vapi markets sub-500ms) is often optimistic; one independent test measured 720ms median and over 1,000ms at the 95th percentile
- !Background noise on real phone lines (shop floors, cars, wind) degrades transcription accuracy far more than it does in demos
- !Forgetting the AI-disclosure line at call start is both a compliance problem and the single most common complaint callers have about these systems
- !Call transfers need a warm handoff with context, or the human picking up has to ask the caller to repeat everything, which defeats the point
- !HIPAA and other regulated use cases need the paid BAA add-on (Vapi charges $2,000/mo for it), which changes the DIY cost math substantially
Compliance & risk reality check
Voice receptionists carry more compliance weight than chat agents because phone calls trigger recording-consent and telecom-specific rules on top of ordinary AI-disclosure rules.
TCPA and FCC AI-voice rules
The FCC's February 2024 Declaratory Ruling (24-17) holds that AI-generated voices count as an "artificial voice" under the TCPA, requiring prior consent for outbound calls; this mainly hits outbound reminder or reactivation calls rather than inbound answering.
Mitigation: If your receptionist also places outbound calls (appointment reminders, callbacks), capture written consent that names AI-generated calls, scrub against the Do Not Call list, and disclose the AI at the start of every outbound call.
Call recording and wiretap law (CIPA-style claims)
Two 2025 California cases, Ambriz v. Google and Taylor v. ConverseNow, both survived motions to dismiss on the theory that an AI vendor capable of using call recordings to train its own models can be an unlawful third-party eavesdropper, even without proving it actually happened.
Mitigation: Play a recording-and-AI notice at the very start of every call, put a no-training clause in every vendor contract covering your call data, and prefer vendors offering zero-retention modes.
AI disclosure on the call itself
California's BOT Act, Utah's AI Policy Act and the EU's Article 50 transparency rule (in force from August 2, 2026) all push toward telling callers they're talking to AI.
Mitigation: State "this is their AI assistant" in the very first line of the greeting, and never claim to be human if a caller asks directly.
PCI DSS if the agent takes payment details by phone
If a caller speaks a card number to the AI to pay a deposit, the recording, transcript and underlying model calls all enter cardholder-data scope under PCI DSS 4.0.1, fully mandatory since March 31, 2025.
Mitigation: Use DTMF keypad capture or a pay-by-SMS link so the card number never reaches the AI, the recording, or the transcript.
All-party consent recording states
California, Florida, Illinois, Pennsylvania, Washington and several other states require all parties to consent to a recorded call, and cross-state calls follow the stricter rule.
Mitigation: Treat every call as if it needs two-party consent regardless of the caller's state, since the opening AI-and-recording disclosure satisfies this in every jurisdiction that matters.
Build vs buy: the real math
3-6 weeks for a custom agency build; a few days for a basic Vapi version
Custom build time
$13K-$25K for an agency build (up to $35K for multi-system or compliance-heavy work); $0-$100/mo for DIY on Vapi
One-time investment
Roughly 1,500-2,000 calls a month, or whenever you need a booking-system integration no packaged SaaS vendor supports
Breakeven vs buying
At 200 calls a month averaging 3 minutes each, Goodcall's Starter plan at $79/mo for 100 unique callers (plus $0.79/extra caller) covers a small business for well under $200/mo. A $13K-$25K custom build amortized over a year runs $1,083-$2,083/mo in year one before you've added a cent of ongoing voice-stack cost, which only makes sense once a packaged tool genuinely can't do the job, such as writing appointments directly into ServiceTitan, a dealership DMS, or a practice-management system no SaaS receptionist integrates with. Past roughly 2,000 calls a month, the per-caller and per-call overage on most SMB tiers starts approaching what the amortized custom build would have cost, which is the volume where it's worth re-running this math.
Skip the DIY — RapidDev builds the production version
A Lovable MVP gets you a demo. Production needs auth that doesn't leak data, AI calls that don't bankrupt you, observability when models drift, and code you can audit. That's what we ship through our AI app development service.
Discovery call (free)
30 minWe map your exact small business AI receptionist use case: who uses it, target volume, AI model choice, integrations, compliance scope. You get a detailed scope document and fixed-price quote within 48 hours.
AI-accelerated build
3-6 weeks for a custom agency build; a few days for a basic Vapi versionOur engineers use Claude Code, Lovable, and custom tooling to ship 3–5x faster than agencies. You see weekly progress in a staging environment — not a black box.
Launch + handoff
1 weekWe deploy to your infrastructure, transfer the GitHub repo, set up CI/CD and monitoring, and train your team. You own 100% of the source code, prompts, and model configurations.
What you get
Timeline
3-6 weeks for a custom agency build; a few days for a basic Vapi version
Investment
$13K-$25K for an agency build (up to $35K for multi-system or compliance-heavy work); $0-$100/mo for DIY on Vapi
vs SaaS
ROI in Roughly 1,500-2,000 calls a month, or whenever you need a booking-system integration no packaged SaaS vendor supports
30-min call. Fixed-price quote within 48 hours. No commitment.
Frequently asked questions
How much does an AI receptionist for small business cost?
Packaged SaaS tools run $29-$349/mo depending on call volume: Goodcall is $79-$299/mo by unique caller, Smith.ai is $0-$500/mo by call. A custom agency build runs $13K-$25K upfront (up to $35K for complex, multi-system work), plus $50-$300/mo in ongoing voice-stack costs. A DIY build on Vapi costs roughly $0.08-$0.13/minute of call time.
How long does it take to set up?
A packaged SaaS receptionist can go live the same day you forward your number, since it's mostly recording a greeting and connecting a calendar. A custom agency build takes 3-6 weeks to wire into your specific booking software and write compliant disclosure and transfer scripts. A basic DIY Vapi version can work within a few days, though getting call quality and transfers production-ready takes 2-4 weeks.
Can RapidDev build this for us?
Yes. RapidDev has shipped 600+ apps and offers a free 30-minute consultation to scope your booking system, call volume and compliance needs before you commit to a custom build.
Will callers know they're talking to AI?
They should hear it in the very first line. Stating "this is their AI assistant" in the greeting satisfies disclosure expectations under California's BOT Act, Utah's AI Policy Act and the EU's Article 50 transparency rule, and it also defuses the awkwardness of a caller figuring it out partway through.
What happens when the receptionist can't handle a call?
A well-built agent recognizes a complex quote, an angry caller, or anything outside its script and warm-transfers the call to a human line with a short spoken summary of what the caller needs, rather than looping them through a scripted dead end.
Can it work with my existing phone number and calendar?
Yes. Most vendors forward or port your existing business number rather than issuing a new one, and connect to Google Calendar, Outlook or Calendly for booking; a custom build can also write directly into vertical software like ServiceTitan or a practice-management system.
Is call recording legal with an AI receptionist?
Yes, as long as you disclose it. Two 2025 California cases let wiretap claims against AI voice vendors proceed past a motion to dismiss specifically because the calls weren't clearly disclosed as recorded and AI-handled, so playing a recording-and-AI notice at the start of every call is both the legal safe harbor and standard practice.
What's the difference between per-minute, per-call and per-caller pricing?
Per-minute pricing (common in DIY voice stacks, around $0.08-$0.15/min) charges for call length, per-call pricing (Smith.ai) charges a flat rate regardless of length, and per-caller pricing (Goodcall) charges once per unique phone number regardless of how many times or how long they call. Per-caller pricing is usually cheapest for businesses with repeat callers.
Want the production version?
- Delivered in 3-6 weeks for a custom agency build; a few days for a basic Vapi version
- You own 100% of the code
- AI cost monitoring built in
30-min call. No commitment.
