AI receptionists in Australia: what works, what flops and what it costs.

What is an AI receptionist, and does it work?

An AI receptionist is a voice agent that answers your business phone, talks with the caller in plain speech, and then does something: takes a message, books a time, answers a common question or puts the call through to a person. It works well for routine, predictable calls. It works badly for upset, urgent or complicated ones.

The technology has moved fast. A good one now sounds natural, copes with interruptions and can look things up while the caller waits. That is the easy part. The hard part is everything around the voice: what it is allowed to say, where the booking goes, who it hands over to and what happens when it does not understand.

We get asked about these on sales calls most weeks, usually by owners who are drowning in calls they cannot answer while they are on the tools. Some of them should buy one. Some of them have already bought one and regretted it. The difference is almost always the type of caller, not the software.

How much does an AI receptionist cost in Australia?

Off-the-shelf AI receptionist apps start at about $49 a month for a set number of minutes, then charge per minute. Voice agent platforms you build on charge by the minute, roughly US$0.07 to US$0.31. A human answering service in Australia starts from about $25 to $35 a month plus GST, before call charges.

The subscription is the small number. What you really pay for is the setup, the testing and the minutes once call volume grows. The figures below come straight from each vendor's pricing page. Several AI vendors show a dollar sign without saying which currency, so confirm at checkout before you budget.

Two things catch people out. First, minutes run out faster than you think: 150 minutes is only about fifty three-minute calls. Second, the cheap plans usually only take messages. Transfers, bookings into a real calendar and anything that touches your job software sit on the higher plans or need a proper build.

What can an AI receptionist do well?

An AI receptionist is at its best on calls that follow a pattern: opening hours, service areas, "can you come out Thursday", "did you get my message", booking a standard appointment. It never misses a call, never has a bad day and can text the caller a link while they are still on the line.

• After-hours and overflow. The calls that currently go to voicemail, which most people hang up on. Even a well-taken message beats a missed call.
• Sorting calls. Working out what the caller needs and putting them through to the right person with a short summary, so nobody has to ask the same questions twice.
• Standard bookings. Fixed-length appointments into a calendar, with a confirmation text. If bookings are your main load, our booking automation work often starts here.
• Following up. Sending a text after every missed or after-hours call so the lead does not ring the next business on Google. That is often worth more than the voice itself, and it is the core of automated lead follow-up.

When does an AI receptionist flop?

It flops when the caller is stressed, the stakes are high or every call is different. Someone broken down on a highway, a tenant with a burst pipe, or a customer chasing a big invoice wants a person who can make a decision. A voice agent that asks them to repeat their address feels like a wall.

The other common failure is a bot that cannot reach the information the caller is asking about. If it cannot see the job, the booking or the invoice, all it can do is take a message, and callers can tell. Check what the receptionist can actually look up before you judge how smart it sounds.

When is a human receptionist still better?

A person is still better when calls are emotional, urgent, high value or varied enough that no script covers them. For many small businesses the best setup is a mix: AI handles after hours and the routine calls, and a person or a human answering service takes the rest.

Human answering services in Australia are not expensive at low volume, and they already know how to calm down an upset caller. If most of your calls are emergencies, start there and use AI behind the scenes instead: summarising the call, logging the job and texting the customer an update.

It is also worth asking whether the phone is really the problem. One owner we spoke to was a Perth plumber working with one apprentice, doing about four hours of admin every night after a full day on the tools. The ask was an AI office assistant. When we listed what that meant, it was templated quotes, "on my way" texts, quote follow-ups and posts from their Google reviews. Answering calls was not the top of the list. Most of the hours were in the paperwork behind the calls.

Should you buy an AI receptionist or build one?

Buy one if you only need calls answered, messages taken and simple bookings made. Build one when the receptionist has to read and update your own systems, like your job software, property system or customer records, because that is where the off-the-shelf apps stop.

A bought app is quick: you load your hours, services and answers to common questions, and it can be answering within the day. A built voice agent takes longer, because the work is connecting it to your data and testing it against real calls. That testing is the part people skip, and it is the part that decides whether customers trust it.

That is the same order one of our clients chose. The Managing Director of Cortes Management Group, a Perth property manager, is building an AI receptionist for tenant and owner calls in stages: the phone line and call handling first, then the live property data behind it. Every step is tested before the next one depends on it.

If you are tempted to build one yourself with a general AI tool, read our guide to ChatGPT agent mode for business first. It is a good picture of what today's agents handle on their own and where they still need a person checking.

How do you set up an AI receptionist without annoying callers?

Start small and test hard. Give it one job, such as after-hours calls, write down exactly what it may and may not say, make it easy to reach a person, and listen to the first fifty recordings yourself before you let it answer during business hours.

1. List your call types. Go through a week of calls and sort them: routine, urgent, sales, existing customers. The routine pile is what the AI should take.
2. Tell callers it is AI. People forgive a voice agent that is upfront. They do not forgive one that pretends.
3. Always offer a way out. "Press 0 or say 'person'" should work from the first second, and urgent words should go straight through.
4. Send a summary, every time. Each call should land in your inbox or job system as a short note, so nothing depends on someone remembering.
5. Review weekly at first. Read the transcripts, fix the answers it got wrong, and only then widen what it handles.

The same habit of reviewing transcripts applies to meetings. If you already use a tool that records calls, our comparison of AI note takers like Fathom, Granola and Otter covers the ones we rely on.