What is an AI receptionist, and what can it handle for an Australian small business?
An AI receptionist is a voice agent that answers phone calls, understands what callers say, responds in natural speech, and takes an action such as booking a slot, taking a message or transferring an urgent call. For a small business it covers the moments when nobody can reach the phone.
Done well, it handles the predictable majority of calls: “Can someone look at my hot water system this week?”, “Do you service Penrith?”, “What time do you open Saturday?”, “I need to move my Thursday appointment.” It collects the caller's name, number, suburb and what they need, checks real availability, and either books or promises a callback within a stated window.
What it should not do is pretend to be a person, quote prices it cannot be sure of, give clinical or safety advice, or keep a distressed caller talking to a machine. The design work is mostly about those boundaries.
The rest of this guide covers the practical questions Australian owners ask: accents, recording laws, bookings, after-hours triage, and whether to build one or subscribe to a service.
How many calls is your business actually missing?
Find out before buying anything: your phone carrier's call records or your phone system's logs show answered, missed and abandoned calls by hour. Two weeks of data is enough to see the pattern.
Look for three things. First, the volume of missed calls during working hours, which shows how often you are on a job or with a client. Second, calls outside hours, which shows after-hours demand. Third, how many missed callers never rang back; if your records show a number calling once and never again, that caller probably found someone else.
Then check what happens now. Does voicemail fill up? Do you return calls the same day? Does anyone listen to messages on weekends? Many owners discover the problem is not the number of missed calls but the delay in returning them.
If the numbers are small, a better voicemail greeting and a missed-call text-back may be all you need. If they are steady, an AI receptionist for small business use starts to make financial sense, and the same data sets the budget for running costs.
Build or buy: custom AI receptionist vs per-minute SaaS
Buy a subscription product when your needs are simple and its built-in integrations cover your booking system. Build a custom AI receptionist when you need specific triage rules, a booking integration the products lack, control over where data goes, or when per-minute pricing at your volume adds up.
Subscription AI receptionists are quick to start: sign up, forward your number, fill in your business details. Their limits show when your rules are unusual (“gas leaks go to Dave's mobile, but only if the caller is in our service area”), when you need bookings written into a system they do not support, or when you want your recordings and transcripts in accounts you control.
A custom build takes a few weeks and needs testing, but it runs on providers you choose, costs only their usage plus optional upkeep once live, and does exactly what you specify. It is also yours to change.
Do the maths at your real volume. Take your monthly call count, multiply by average call length, and price it on the subscription you are considering. Compare with a build from US$600 plus estimated running costs over two years. Neither always wins.
AI receptionist or human virtual receptionist: which should a small business choose?
Choose a human virtual receptionist when warmth, judgement and complex conversations matter more than cost, and an AI receptionist when calls are mostly predictable and you want them answered instantly at any hour. Plenty of businesses use both.
Human answering services are good at reassurance. A caller whose ceiling has just collapsed wants a calm person, and a trained receptionist can handle a rambling story without losing the thread. Their limits are cost per call, availability at 3 am, and the fact that most can only take a message rather than book directly into your system.
An AI receptionist never puts a caller on hold, answers ten calls at once during a storm, and writes a clean summary every time. It struggles with genuinely unusual conversations, which is why the hand-off rules exist.
A common hybrid for Australian trades and clinics: the AI answers first, books the routine work and captures details, and anything emotional, complex or urgent goes straight to a person, whether that is you, a staff member on call or a human answering service you already pay for. We can build the transfer to whichever human option you use.
Does an AI receptionist understand Australian accents and place names?
Modern speech recognition handles general Australian English well, but it still stumbles on unusual suburb names, trade jargon, noisy lines and some accents. The fix is testing with your real call types, adding vocabulary hints, and always confirming important details back to the caller.
Suburbs such as Woolloomooloo, Mooloolaba, Kurri Kurri or Koo Wee Rup, street names borrowed from Aboriginal languages, and trade terms like “TPR valve”, “RCD tripping” or “sarking” are exactly where generic systems fail. We load your service-area suburbs and common job terms into the recognition and the agent's instructions, so it expects them.
Confirmation is the second safety net. The agent reads back the phone number digit by digit, repeats the suburb, and spells out anything unclear. If it still cannot understand after two attempts, it apologises and offers a callback from a person instead of guessing. Callers on a building site or in a car get the same fallback.
Before launch we run test calls covering your common job types, varied speakers and background noise, and we review real transcripts in the first weeks to catch patterns. Accuracy is measured, not assumed.
Call recording and surveillance laws: what an AI receptionist must announce
Because recording and surveillance laws differ between states and territories, the simplest safe design is for the AI receptionist to say at the very start of each call that it is an automated assistant and whether the call is recorded or transcribed. Get your own lawyer to confirm the wording for where you operate.
The differences are real. New South Wales's Surveillance Devices Act 2007 generally prohibits recording a private conversation with a listening device without the consent of the principal parties, subject to narrow exceptions. Victoria's Surveillance Devices Act 1999 is framed around recording private conversations to which the person is not a party. Other states and territories have their own laws, and federal telecommunications law also applies to phone calls. Callers may be in a different state from you.
We therefore build the notice in by default: a short, plain greeting such as “You're speaking with the automated assistant for Smith Plumbing; this call is recorded so we can follow up.” Callers who object can ask for a callback, and the agent stops and takes a number only. You can also choose to keep no audio at all and store only a text summary. We do not give legal advice; your lawyer confirms the approach.
Privacy, AI disclosure and caller data
Treat call transcripts and summaries as personal information: collect only what you need, store it securely, and tell callers they are talking to an AI. The OAIC's guidance on commercially available AI products recommends that public-facing AI tools be clearly identified as such.
If your business is covered by the Privacy Act, the Australian Privacy Principles apply to what the receptionist collects. The OAIC says most businesses with annual turnover of AUD 3 million or less are not covered, but health service providers are covered regardless of turnover, which captures many clinics, physios and allied health practices. Your adviser should confirm your position and your privacy policy wording.
On the build side we keep things narrow. The agent asks only for what the booking needs, not date of birth or Medicare numbers unless your process genuinely requires them and your adviser agrees. Transcripts sit in accounts you own, with retention you set. AI providers are business API accounts in your name, and we state which providers and regions process each call in the written scope. See the OAIC's AI product guidance for the regulator's view.
Booking calls into ServiceM8, Cliniko or Google Calendar
An AI receptionist is most useful when it can see real availability and write the booking or enquiry straight into your system. All three common Australian choices can be connected, each in its own way.
ServiceM8 for trades
ServiceM8's REST API works with jobs, clients (called Company in the API) and job activities for scheduled bookings. The agent can create a job with the caller's details and description, and either propose a time from your rules or leave it as an unscheduled enquiry for the office to slot in.
Cliniko for clinics
Cliniko publishes an API with appointment, available-times and booking resources. The agent can offer genuine open slots for the right practitioner and appointment type, then create the booking, following the rules your practice sets about new versus existing patients.
Google Calendar for everyone else
For sole traders and small studios, Google Calendar works well: the agent checks free time within your working hours, adds a buffer for travel, and creates an event with the caller's details in the description.
Many owners prefer “request and confirm” rather than instant booking for the first month: the agent collects a preferred time, you confirm by text. Once you trust it, switch to direct booking. Deeper job-system work sits under custom software development.
After-hours triage: what counts as an emergency?
You define emergencies, not the AI. We write your triage rules as a short decision list, and the agent follows it: genuine emergencies are escalated immediately, urgent jobs are booked first thing, and routine enquiries get a normal callback.
For a plumber, “water pouring through the ceiling” and “smell of gas” escalate; a dripping tap waits. For an electrician, sparking, burning smells or total loss of power escalate. For a clinic, the agent never triages clinically: anyone describing a medical emergency is told to hang up and call Triple Zero (000), and anything else is a message or booking. For a locksmith, a person locked out at night may escalate while a lock change waits.
Escalation itself has options: warm transfer to your mobile, a loud text with the summary, or a call to an on-call roster in order. If nobody answers the escalation, the agent tells the caller honestly what will happen next rather than promising a time you cannot meet.
- Write three lists: escalate now, book first thing, normal callback
- Name who receives escalations, in what order
- Decide what the agent says if no one picks up
- Set a safety line for anything life-threatening: call 000
Handing calls to a human without annoying the caller
A good AI receptionist makes it easy to reach a person: callers who ask for a human are transferred or offered a callback immediately, not argued with. Frustration with voice bots mostly comes from being trapped.
We build three hand-off paths. Warm transfer connects the caller to your mobile with a whispered summary before you are connected, so you do not ask them to repeat themselves. Callback capture takes a number and best time when no one is free. Message relay sends a summary by SMS and email within a minute of the call ending.
The agent also hands off by itself when it detects it is not helping: repeated misunderstandings, an angry tone, a complaint, or a topic outside its knowledge. Those rules are written into the call flow and tested.
Writing the AI receptionist's script and knowledge base
The agent answers from a short knowledge base you approve: services, service areas, hours, what you do not do, and standard answers. It does not browse the internet or improvise about your business.
We draft the knowledge base with you from your website, a call with you and a few recordings of typical calls if you have them. It includes a greeting, the recording and AI notice, the questions to ask for each job type, suburbs you cover and ones you do not, policies such as call-out charges described in words rather than figures unless you want exact prices spoken, and phrases to avoid.
Keep it current. When you stop offering a service or change hours, the knowledge base changes the same day. During the free tuning period we update it for you; afterwards, you can edit simple text sections yourself.
How much does an AI receptionist for small business cost in Australia?
A custom AI receptionist starts from US$600 to build. Running costs depend on call minutes and are paid to providers in your name; subscription products instead charge per minute or per plan, and prices vary widely between them.
Build cost rises with the number of call types, the booking integration (Google Calendar is simpler than a clinic system with practitioner rules), multi-location routing, escalation rosters and how much testing your callers' variety demands. Running cost has four parts: the phone number and call minutes, speech recognition, synthesised voice and the AI model. Longer calls cost more, which is one reason good agents keep calls short and focused.
To compare fairly, take your real monthly call count and average length from the missed-call data, price it on each subscription option, and set that against the build plus estimated usage over two years. We put that comparison in the written quote.
What an AI receptionist should never do
Some things stay with people, whatever the technology can do. Writing these into the design protects your customers and your reputation.
- Pretend to be a human or hide that the call is recorded
- Give medical, legal, electrical-safety or gas-safety advice
- Quote firm prices for work nobody has seen
- Promise arrival times the roster cannot meet
- Take card payments over the phone
- Collect sensitive details the booking does not need
- Keep a distressed or angry caller from reaching a person
We encode each of these as a rule and test it with deliberately awkward calls before launch.
How to tell whether your AI receptionist is working
Measure booked jobs and returned calls, not just answered calls. An agent that answers everything but books nothing is an expensive voicemail.
Each week of the first month we review four numbers with you: calls answered by the agent, bookings or qualified enquiries created, calls handed to a person and why, and calls where the caller hung up early. We also read a sample of transcripts for misunderstandings. Early fixes are usually small: a suburb it mishears, a question it asks in the wrong order, a service it does not know you offer.
Compare against the missed-call baseline you collected at the start. The goal is fewer callers disappearing and faster responses, and you should see both in your own records.
How the build works with a team in India
You will speak to us in your afternoon, our morning: India is 4.5 hours behind AEST, 5.5 hours behind AEDT, and 2.5 hours behind Perth. Test calls can happen at any hour, because the agent works around the clock once it is live.
Week one is a 45-minute call about your calls: types, triage rules, booking system and escalation contacts. We send a written scope listing telephony, speech and AI providers and their regions, the recording notice wording for your lawyer to check, and the running-cost estimate. Week two, after your written approval, you create the provider accounts in your business name and invite us. By the end of week two you can ring a test number and hear the agent. Weeks three and four are testing, tuning and forwarding your real number.
Quotes are in USD, paid by Wise, wire or PayPal, with invoices from India. We answer WhatsApp seven days a week and keep scope changes in writing.
Worked example: a Townsville air-conditioning installer after hours
This example is hypothetical. Imagine a Townsville air-conditioning business with three installers and no office staff. In summer the owner misses calls all day on rooftops, and after-hours calls about units failing in the heat go to a full voicemail box.
The build forwards unanswered and after-hours calls to an AI receptionist. It opens with the AI and recording notice, asks for name, number, suburb, the unit type and the problem, and checks the suburb against the service area. Breakdowns for elderly or medically vulnerable callers during a heatwave escalate to the owner's mobile under his rules; routine services become unscheduled ServiceM8 jobs with the summary attached; quote requests get a callback slot. The owner receives a text within a minute of each call.
The written quote would start from US$600 for the build, with ServiceM8 integration and the escalation roster listed, and running costs estimated from his summer call counts. After four weeks he compares his carrier's missed-call records with the agent's booking log to decide whether to keep direct booking on.
AI receptionist checklist for Australian small businesses
Gather this before you speak to any provider, whether you buy a subscription or commission a build.
- Two weeks of call records: answered, missed, after hours
- Your booking system and who has admin access
- Call types and the questions each needs answered
- Escalate / book / callback triage lists
- Escalation contacts and the order to try them
- Service-area suburbs and ones you do not cover
- Recording and AI notice wording checked by your lawyer
- Privacy Act coverage confirmed with your adviser
- Who reviews transcripts weekly in the first month