AI receptionist for small business: calls, bookings and handoffs

The phone rings while you are with a customer, and you have to choose who gets your attention. An AI receptionist removes that choice for routine calls. It picks up on the first ring, answers the questions you hear ten times a day, and puts real bookings in your diary, so the calls that reach you are the ones that need you.

Callers still want to talk. In Hiya’s State of the Call 2025 survey, which covered more than 12,000 consumers in the US, UK, Canada, France, Germany and Spain, three times as many consumers said they prefer voice over text when dealing with retail businesses. This guide covers what an AI receptionist does, where it stops, which rules apply, and how to launch one with a script and a pass/fail checklist.

What is an AI receptionist?

An AI receptionist is software that answers inbound calls in a natural voice, understands what the caller wants, and completes the simple jobs a front desk does. It runs on your existing number, usually through call forwarding, around the clock. It is not a phone menu: callers speak normally and can interrupt mid-sentence.

Here is how it compares with three older options.

Option	Answers questions	Books appointments	Works after hours	Typical weak spot
Voicemail	No	No	Takes messages only	Nothing happens until someone calls back
Phone menu (IVR)	Fixed recordings	Rarely	Yes	Rigid options frustrate callers with unusual requests
Human answering service	From a script	Sometimes	Yes, at extra cost	Cost rises with every call or minute
AI receptionist	From your approved answers	Yes, in your live calendar	Yes	Needs clear rules and a human fallback

The difference that matters most is the calendar. A message-taker leaves you a list of callbacks; an AI receptionist ends the call with the appointment made and confirmed.

How does an AI receptionist handle a call?

An AI receptionist handles a call in six steps, and each one is something you configure, not something it improvises.

1. Greeting and disclosure. It answers with your business name, says it is an AI assistant, and gives any recording notice you need.
2. Intent. It works out what the caller wants: a price, opening hours, a booking, a specific person, or something else.
3. Answer. It replies only from the FAQs and documents you approved. If the answer is not there, it says so and offers a callback or transfer.
4. Booking. It reads live availability, offers slots that follow your rules (length, buffers, notice period), books, and reads the details back.
5. Transfer. Callers who ask for a person, or whose request is out of scope, go to the number you chose. If nobody picks up, it takes a structured message.
6. Record. A transcript, a two-line summary and the outcome go into your CRM or inbox, so nothing depends on someone remembering to write a note.

Step 3 is where most quality problems start: a receptionist allowed to guess will eventually quote a wrong price. Document-based chatbots use the same restriction, explained in how a chatbot answers from your own documents.

What should an AI receptionist handle, and what should stay with people?

An AI receptionist should handle calls with a known answer and a clear end point, and pass everything that needs judgment to a person. A useful test: if a new member of staff could handle the call on day one with a printed sheet, the AI can too.

Good jobs for the AI

• Opening hours, location, parking and directions
• Standard prices and what a service includes
• New bookings, reschedules and cancellations inside your rules
• Order or job status, when it can look the status up
• Detailed after-hours messages with the caller’s number and reason

Jobs that should reach a person

• Complaints, refunds outside policy and anything emotional
• Quotes that depend on a site visit or a diagnosis
• Medical, legal or financial questions: the AI books the consultation, and a qualified person gives the advice
• Existing clients who ask for someone by name
• Anyone who says they want a human, at any point in the call

That last rule protects your reputation. In a Gartner survey of 5,728 customers, 64% said they would prefer companies did not use AI for customer service, and the top concern was that it would get harder to reach a person. Transferring on request answers that concern directly.

What rules apply when AI answers your phone?

Three sets of rules matter for most US and European businesses: AI disclosure, call-recording consent and, if the system also calls people back, robocall consent. None of them stops you using an AI receptionist, but each one shapes the script.

Telling callers they are talking to AI

In the EU, Article 50 of the AI Act requires AI systems that interact with people to be designed so people are informed they are interacting with an AI system, unless that is obvious, and at the latest at the first interaction. The European Commission confirms that Article 50 applies from 2 August 2026. Outside the EU, a one-line disclosure in the greeting is still the honest default, and it costs about three seconds per call.

Recording and transcribing calls

US federal law allows recording when one party to the call has given prior consent, but the Digital Media Law Project lists about ten states, including California and Florida, that require the consent of every party. Because callers can be anywhere, the practical answer is a short recording notice in the greeting on every call. In Europe, recordings and transcripts are personal data, so your privacy notice should cover them.

Calling people back

Inbound answering is the simple case. If the system also places reminder or follow-up calls, note that the FCC ruled in February 2024 that AI-generated voices count as artificial voices under the TCPA. Those calls need the called person’s prior express consent, and prior express written consent for marketing calls, unless an exemption applies. This guide is not legal advice; check your own obligations with counsel.

How to set up an AI receptionist in 8 steps

You can prepare most of an AI receptionist before anyone builds anything. These steps take a few hours spread over a week, and they decide most of the quality you get.

1. Pull 30 days of call history. Export the log from your phone system. Count calls by hour of day and note how many went unanswered or to voicemail.
2. Sort call reasons. Listen to 30 to 50 recent calls, or ask whoever answers the phone, and tally the reasons. A short list usually covers most calls.
3. Write the answer sheet. For each common question, write the answer exactly as you want it said, in plain spoken English. Add what the AI must never say, such as discounts or diagnoses.
4. Write your booking rules. Appointment types, lengths, buffers, minimum notice, which staff member takes which type, and how far ahead people can book.
5. Draw the transfer map. For each call type that needs a person, name who takes it, on which number, during which hours, and what happens if they do not answer.
6. Agree the greeting. Business name, AI disclosure, recording notice and one sentence on what the caller can do. Keep it under 15 seconds.
7. Test with realistic calls. Run at least 50 test calls, including accents, background noise, callers who change their mind, and people who ask for a human straight away. Score them with the checklist below.
8. Start narrow, then widen. Go live on after-hours and overflow calls first. Review transcripts weekly for a month, fix the answers, then decide whether it takes the main line.

Step 1 also gives you a baseline, so you judge the receptionist on calls caught and bookings made, not on how impressive it sounds. The numbers to watch afterwards are in our guide to customer support metrics after adding AI.

A sample greeting and transfer map

Example: a two-van plumbing firm. Its greeting, read at a normal pace, takes about 12 seconds:

“Thanks for calling Smith Plumbing. I’m the AI assistant, and this call is recorded. I can book a quote visit, answer questions about our services, or put you through to the team. How can I help?”

Its transfer map fits in one table, and the AI follows it exactly:

Call type	Goes to	When	If nobody answers
Emergency: leak, no water, no heating	On-call mobile	Any time	Takes details and texts them to the on-call plumber
Question about a job in progress	Office line	Mon to Fri, 8 am to 6 pm	Structured message for a callback next working day
Complaint	Owner	Office hours	Message flagged urgent in the CRM
“Can I speak to a person?”	Office line	Office hours	Message, and the caller is told when to expect a call

The go-live checklist: pass or fail

Run the 50 test calls against these checks. Any single fail means fix and re-test before real callers hear it.

• Disclosure and recording notice heard on every test call.
• Zero wrong prices, invented services or promised discounts.
• “I want a person” reaches the transfer within one turn, 10 times out of 10.
• Every test booking lands in the calendar with the right length, buffer and staff member.
• An unanswered transfer falls back to a message every time, never to silence.
• Every call leaves a transcript and summary in the CRM.

If many of the same questions also arrive by text, the same answer sheet can run a messaging channel: see WhatsApp AI agents for business. Speed matters most for new enquiries, because a caller who reaches voicemail often rings the next business on the list; we look at that in speed to lead and automating first contact.

How much does an AI receptionist cost to run?

An AI receptionist costs a one-time build, ongoing usage that is mostly call minutes billed by your telephony and AI providers, and optional maintenance. Call length drives the bill more than call count. The arithmetic, with published provider prices, is in how much a voice AI agent costs to build and run.

How Upmark Labs can help

We build AI receptionists at a fixed price, with the limits shown on a Scope Card before you pay. It is built in your own accounts, so your number and call data stay yours.

Our AI Voice Receptionist starts from $4,900. In scope: one phone number, up to two connected systems such as your calendar and CRM, up to 50 documents or one FAQ set as its knowledge, up to five call flows plus live transfer, and one language. We run 50 test calls and send you the transcripts, you get two rounds of changes and a recorded 60-minute handoff session, and 30 days of support follow launch. Standard delivery is 20–30 working days, with paid Express and Rush options at checkout.

You provide a telephony account in your name, your scripts and FAQs, the transfer numbers, calendar and CRM access shared by invite or a key you control, your recording disclosure wording and one person who can approve decisions, all within 72 hours of ordering.

It is not the right product if most calls need expert judgment, if you need more than five call flows, or if you mainly make outbound calls; the product page points to the better fit for each. Not sure your callers will accept an AI voice? The Voice Agent Pilot runs one call flow on your real after-hours or overflow calls for 14 days, from $1,625, and the fee is credited if you order a full voice build within 30 days of the pilot ending. Booking-only lines and industry versions are in the Voice AI store.

Frequently asked questions

If you already know your top call reasons, the next step is the AI Voice Receptionist page, where the Scope Card shows exactly what is included. If you would rather talk it through first, book a free expert consultation and bring your last month of call logs.