AI receptionist: the maths on when it pays for itself in an AU small business

The average AU service business misses between eighteen and thirty percent of incoming calls. Most of those callers do not leave a message and most never call back. They call the next business on the Google result.

I run three petrol stations and I sit on the build side of a dental clinic site. I see the missed-call problem from both angles, the operator watching the after-hours leads slip away and the developer building the tool that catches them. The maths is not theoretical.

This post is the honest break-even calculation on AI receptionists for AU SMBs in 2026. Who they work for. Who they do not. And the specific call volume threshold where they cross from "nice idea" to "obvious yes."

What an AI receptionist actually does in 2026

An AI receptionist is a voice agent that answers your business phone. Not a chatbot on your website. A real phone call, picked up on the second ring, twenty-four hours a day.

The 2026 version handles roughly five jobs. It answers basic questions like opening hours, location, parking, and pricing. It checks live stock or live calendar availability if you wire it to your systems. It books simple appointments straight into your booking tool. It escalates urgent calls to a staff mobile during business hours with a summary text. And it captures every caller as a contact in your CRM, including the ones who hang up before booking.

The voice quality crossed the credibility line about eighteen months ago. Customers can tell it is AI on the first sentence, but the conversation flows, and the friction to keep talking is now lower than the friction to hang up and try the next business. That was not true in 2023. It is true now.

For context on where this sits in the broader AI tooling landscape for AU small business, the phone receptionist is the highest-ROI move on the list for any business where the phone actually rings.

The missed-call problem, in numbers

Here are the rough miss rates I have seen across AU service businesses over the last year, taken from logs, not vendor decks.

• Plumbers and electricians, twenty to twenty-five percent missed during business hours, sixty to seventy percent after hours
• Dental clinics, fifteen to twenty percent missed during business hours, ninety percent plus after hours
• Bottle shops, twenty-five to thirty percent missed across the day, almost all after seven pm
• Vets, ten to fifteen percent missed during hours, eighty percent after hours
• Hair salons, twenty percent missed during peak booking windows

The number that matters is not the miss rate. It is the dollar cost of each missed call.

The maths is straightforward. Average customer value, multiplied by your call-to-booking conversion rate, multiplied by the number of missed calls. A plumber with a four hundred dollar average job and a forty percent call-to-booking rate is leaking one hundred sixty dollars in expected value for every missed call. Miss four a day and that is six hundred forty dollars in leaked revenue daily, roughly fifteen thousand a month.

Most operators have never run this calculation. The number is always larger than they expect.

Who AI receptionists work best for

The fit is sharpest where the calls are high volume and the questions are predictable.

Strong fit. Dental and medical clinics, plumbers and electricians, locksmiths, bottle shops, vets, hair and beauty salons, mobile mechanics, cleaning services. Any business where ninety percent of inbound calls ask one of about ten standard questions and the booking flow is short.

Weaker fit. Bespoke consulting where every conversation is genuinely unique. High-touch financial advice. Custom builders and architects where the relationship discovery on call one is the actual product. Medical and legal where compliance constraints rule out automated intake without a human in the loop.

The test is simple. Sit with your phone log for a week and count how many of your inbound calls are the same five questions in different words. If it is more than seventy percent, you are in scope. If it is under thirty percent, you are not.

One AU dental clinic we looked at had ninety-two percent of inbound calls falling into four categories. Booking, rescheduling, fee enquiry, and emergency. That clinic was always going to win with an AI receptionist. The custom architect down the road would not.

The maths: when does it pay back

Here is a concrete example. Single-location AU dental clinic doing roughly fifty thousand dollars a month in revenue. Four hundred dollar average new-patient value. Forty calls a day inbound, of which four are missed during business hours and another six come in after hours and go to voicemail. Of the ten missed and after-hours calls, the receptionist confirms that roughly six would have booked if answered.

That is six lost bookings a day at four hundred dollars expected value each, which is twenty-four hundred dollars a day in leaked revenue. Even at a conservative thirty percent conversion on missed calls actually being recovered by an AI receptionist, you are catching seven hundred twenty dollars a day. Across a twenty-two-day month, that is fifteen thousand eight hundred forty dollars in recovered revenue.

The AI receptionist cost for this volume runs between three hundred and five hundred dollars a month all-in, depending on whether you go productised or build it yourself.

Net positive in the first week. Net positive at any plausible recovery rate above five percent. The break-even is so far below the realistic outcome that the decision is not even close.

The dental clinic example is the cleanest because the average customer value is high. A plumber with a three hundred dollar callout and a fifty percent conversion gets to the same answer. A bottle shop with twenty dollar baskets needs a lot more call volume to justify the spend, which is exactly the next point.

The maths when it does not work

The AI receptionist is not a universal yes. Three patterns where the maths breaks.

Low call volume. If you take fewer than twenty calls a day across all channels, the missed-call dollar value rarely clears the monthly receptionist cost. A clean voicemail with a sixty-second callback promise probably outperforms an AI agent at that volume. The receptionist's strength is at scale, not at five calls a day.

Low average customer value. A cafe taking five dollar coffee orders does not need an AI phone agent. The maths only works when each recovered call is worth real money. Twenty dollars and under, skip it. Above one hundred dollars per converted call, run the numbers.

Bespoke conversations where discovery is the product. Custom home builders, family lawyers, M&A advisors, executive coaches. If your first call is forty minutes of listening and the booking flow does not exist, the AI receptionist is the wrong tool. A human picking up matters more than a tool that books slots that do not exist.

The honest test, again, is the phone log. If your inbound calls are mostly bookings, mostly enquiries that fit a template, and mostly worth more than fifty dollars in expected value each, you are in scope. Otherwise spend the money somewhere else.

What to ask before signing up

Most AI receptionist platforms in the AU market look similar from the outside and behave very differently under load. The five questions that separate the real ones from the demos.

One. Does it use my live data or just script answers? A receptionist that reads from a static FAQ doc is a chatbot with a phone number. A receptionist wired into your live calendar, stock, and CRM is actually useful. Ask for proof of integration, not screenshots of marketing pages.

Two. What language model is underneath? This matters less than it did, but it still matters. GPT-4 class and Claude class models are the floor in 2026. If the answer is "our proprietary model," ask for the latency numbers and a live demo on your own phone number.

Three. How does it pass calls to staff? With full context as a summary text or just a ringing transfer with no notes. The first is gold. The second is friction your staff will hate by week two.

Four. Where does the call data live and do I own it? If the answer is anything other than "in your CRM, in your control, exportable," walk. Your customer call data is yours.

Five. Monthly cost vs per minute and cancellation terms. A flat monthly is usually cleaner than per-minute pricing for predictable budgeting, but per-minute can win for very low volume. Either way, the contract should be month-to-month or quarterly. Anyone asking for twelve months upfront is hedging against their own churn.

If the platform cannot answer those five in writing, the platform is not ready.

DIY vs platform

The build vs buy question shows up here too, and it is the same logic as the broader custom software vs SaaS framework.

A DIY build with Twilio plus a custom GPT integration plus your own booking logic runs roughly six to twelve thousand dollars upfront and one hundred fifty to three hundred a month in usage. You get total control, full data ownership, and the ability to wire it directly into your stack. You also own the maintenance.

A productised platform like Recepta, an AI receptionist Voltari built for AU service businesses, runs roughly three to five hundred a month with no upfront. You give up some control, you do not own the underlying integration, but you also do not have to think about it.

For most AU SMBs under five million in revenue, the productised platform is the right call. For multi-location operators or businesses where the call flow is genuinely distinctive, the custom build pays back inside a year. If you are not sure which side of that line you sit on, that is the conversation a fractional CTO can shortcut in an hour, not a vendor sales call.

The bottom line

This is not an "AI is the future" pitch. It is a maths question with a specific answer.

If your business takes more than thirty calls a day, your average customer is worth more than one hundred dollars, and you are missing one in five, the AI receptionist break-even is fast and the upside is real. The dental clinic example in this post recovers fifteen thousand a month against a five hundred dollar cost. That is not a marginal call.

If you take five calls a day from your phone in your pocket, your average customer is worth thirty dollars, and you have never missed a call you cared about, skip it. The maths does not work and the spend buys you nothing.

For the businesses in the middle, where the answer is not obvious from the phone log, the right move is a two-week trial on a productised platform with the missed-call recovery rate measured directly. The data settles the question faster than any spreadsheet.

For a broader view of how AI integrations fit into a real custom build, the Voltari services list covers the Operator tier where this work usually lives. And the case studies on the homepage, including Recepta itself, are the live examples in production right now.

Run the phone log. Do the maths. The answer is usually clearer than you think.