AI Receptionist for Small Business: The Complete Guide
Your phone rang eleven times yesterday while you were under a house, on a ladder or with a patient. Four of those callers left a voicemail. The other seven rang the next business on the list. You will never know what those seven were worth, which is exactly why the problem survives year after year.
An AI receptionist for small business is a system that picks up those calls, and the website chats and Instagram messages arriving at the same time, and does something useful with them instead of taking a name and number. It answers the question if it can, books the job if the calendar allows it, and passes the call to you if the rules say a human should handle it.
I build these for New Zealand service businesses, so this guide is written from the setup side rather than the brochure side. It covers what the system does end to end, where it fails, what the pricing bands look like, and the handoff rules that decide whether customers like it or hate it.
What an AI receptionist actually does on a call#
The call arrives on your existing number. Your telephony provider forwards it to the AI agent, which answers within a ring or two and greets the caller with your business name. Nothing about the caller's experience changes up to that point.
From there the agent runs a short conversation with a purpose. It works out why the person is calling, checks that against what it has been taught about your business, and either answers, books or escalates. A caller asking whether you service Papakura gets a yes or no in four seconds. A caller asking for a rough price on a hot water cylinder replacement gets your actual price band and a booking offer. A caller who is upset about a job you did last week gets handed to you with a note explaining why.
The agent is only as good as what you feed it. A system trained on your service list, your prices, your suburbs and the twenty questions you answer every week performs well. One that has been pointed at your website and left to guess produces vague answers that make customers less confident than before they called. I go through the source material in detail in what to feed an AI agent about your business.
The same agent normally covers more than the phone. Website chat, Instagram DMs and WhatsApp all flow into the same brain with the same rules, so a customer gets a consistent answer whether they rang you or messaged you at 10pm.
The calls it handles well, and the ones it should never take#
An AI receptionist is strong on volume and repetition. Hours, service area, pricing bands, appointment availability, order status, directions, whether you take EFTPOS, when you can come out. These questions make up the bulk of most inbound call volume, and they are the ones that pull you off a job for ninety seconds at a time.
It is also strong on triage. Sorting an emergency callout from a next-week quote request is a question of asking two or three things in order and applying your rules. The agent does not get tired at 4:45pm on a Friday and it does not skip a question because it recognises the caller's voice.
Where it should stand back is anything with money, emotion or ambiguity at stake. A complaint about work already done needs a person. A negotiation on a large quote needs a person. A caller who is distressed needs a person immediately, without being asked to explain the problem twice. Those are rules you set at configuration time, not something you hope the model works out.
In appointment-based businesses the triage job shifts towards reschedules and gaps in the book, which is a different shape of value again and one I cover in AI for dental clinics.
The failure modes are worth knowing before you buy. The agent will occasionally mishear a street name on a bad mobile connection. It will answer confidently from stale information if you changed your prices and never updated the source. It will follow a booking rule to an absurd conclusion if the rule has a hole in it, which is why I write about the specific ways AI booking agents break rather than pretending booking is solved.
What it costs, and what drives the number up#
Pricing in this market breaks into three shapes, and the difference between them matters more than the headline figure.
| Shape | What you should expect to pay | What you get |
|---|---|---|
| Self-serve tool | Roughly $50 to $250 a month | You configure it yourself, chat or basic voice, no integration into your calendar or job software |
| Configured system | Roughly $400 to $1,500 a month plus setup | Built around your services and rules, calendar and CRM connected, escalation configured, someone maintains it |
| Custom build | $8,000 upward as a project, plus running costs | Unusual workflows, integration with older job management software, data that has to stay in specific systems |
Voice costs more than chat because voice bills by the minute on top of the platform fee. A business taking 200 calls a month at an average of two minutes is buying around 400 minutes of voice processing, and per-minute rates sit in the cents rather than the dollars, but that line grows directly with your call volume in a way chat does not.
Prices vary a great deal by vendor and by scope, and the quote you get should be read carefully for setup fees, integration charges and what happens when you want to change the script in month four. I break the bands down further, including the fees that get left off quotes, in what an AI chatbot costs in New Zealand.
Whether any of that is worth paying depends on a number you probably have not calculated. Before you compare vendors, work out what your missed calls are costing you using your own call log and job values. If the answer is a few hundred dollars a month, a self-serve tool is the honest recommendation. If the answer is five figures a year, the configured band pays for itself quickly.
Where a human receptionist still wins#
I am not going to tell you AI beats a person at answering the phone. A good receptionist who knows your regulars, hears the tone in someone's voice and can make a judgment call about squeezing a job in is worth more than any system I can build. What a receptionist cannot do is answer at 9pm on a Sunday, hold twelve conversations at once during a storm callout surge, or work the week they are on leave.
Most businesses that get this right end up with a split rather than a replacement. The AI covers after hours, overflow when the line is busy, and the repetitive daytime questions that were eating the receptionist's attention. The person covers the conversations that need judgment and the customers who matter most. I go through the cost and quality trade-offs properly in AI vs human receptionist, including the situations where hiring is clearly the better buy.
That split also answers the question business owners usually ask second: does this mean I make someone redundant? In practice the receptionist stops being an answering machine and starts doing the work that was always getting deferred, like following up quotes that never got a response.
The handoff is what customers actually judge#
Every complaint I have heard about an AI answering a business phone comes back to the same failure. The customer wanted a human, and the system made that difficult.
Three things fix it. The agent should offer a human whenever a caller asks, without requiring the magic word. It should detect frustration and escalate before the caller has to demand it. And when it transfers, it should pass the full conversation across so the customer never repeats what they just said. That last one is the difference between a transfer that feels seamless and one that feels like starting over. The mechanics of doing it properly are covered in getting the AI handoff to a human right.
Out of hours, the handoff becomes a promise instead of a transfer. The agent takes the detail, tells the caller exactly when someone will be in touch, and puts the job in front of you first thing. A specific commitment beats a vague one, and the agent should never invent a callback time your calendar cannot support.
Setting one up without breaking anything#
The sequence that works looks like this.
- Record your current baseline. Pull last month's missed calls, your average job value and your rough close rate on enquiries. Without this you cannot prove the system worked.
- Write down the twenty questions you answer every week, with the answer you would actually give. This is the single highest-value hour in the project.
- Set the escalation rules. Name the situations where the agent must fetch a human, and be generous with them at the start.
- Run it in shadow mode for a week. Let it handle website chat only, or let it answer overflow calls while you review every transcript.
- Turn on after-hours coverage first, then daytime overflow, then anything else.
Most businesses want to launch at step five. Skipping the shadow week is how you discover a bad answer from a customer rather than from a transcript.
Expect to spend a few hours across the first fortnight reading conversations and correcting the agent. After that, the maintenance is a monthly skim of the transcripts and an update whenever your prices or services change.
What to do this week#
Look at your phone log for the last seven days and count the calls that went unanswered, including the ones that rang out while you were on another call. Then check your website chat and Instagram inbox for messages you never replied to. That combined number, multiplied by your close rate and your average job, is the size of the problem an AI receptionist is being asked to solve.
If the number is small, fix your voicemail message and move on. If it is large, the next question is whether the after-hours share is the biggest slice, because that is usually where a system like this pays back first. The pattern behind that is covered in what to do about after-hours enquiries, and the trades-specific version of the same problem sits in AI customer service for tradies.
Common questions
What does an AI receptionist actually do?
It answers inbound calls, website chats and social messages, identifies what the caller wants, answers common questions from your own information, books or reschedules appointments in your calendar, and passes anything outside its rules to a person with the full conversation attached.
Can callers tell they are talking to AI?
Many can, and the good systems say so at the start rather than pretending otherwise. Callers rarely mind an AI that answers immediately and knows the answer. They mind an AI that stalls, loops or refuses to connect them to a human.
How long does it take to set up an AI receptionist?
A basic setup answering questions and taking messages can run within a week. Adding calendar booking, customer records and escalation rules usually pushes it to two or four weeks, most of which is you deciding the rules rather than anyone writing code.