Parts of it, probably, and nobody selling you a receptionist knows which parts. Valuations and capability are different things: a price can fall without a working phone receptionist stopping work, and a price can rise without one getting any better. Neither is a reason to leave your phone ringing out. Judge the specific job you want done, on a line you can ring, rather than the sector it belongs to.

What if the tool I build on disappears?

A fair concern, and the mitigation is unglamorous: own the record of every call, keep the brief and the integrations written down, and prefer a setup where the rules belong to you and the underlying model can be swapped. On Durbacti's own line every call lands in a database Durbacti controls, the brief is a document, and the voice, transcriber and model are each one setting. Lock-in is a design choice, usually someone else's.

Which job should a small business hand to AI first?

Start with the phone outside your hours, if it rings then. Answering a call, booking a time and taking the caller's details are jobs a new starter could do on day one, so they are the jobs an AI receptionist is best placed to take, and the record of each call shows you what happened. Run it for a week, read the records, then widen the brief to the calls it handled well. Leave anything that needs a judgement about your business with a person until the record says otherwise.

Is AI Here to Stay? Judge the Job, Not the Sector

Ask whether AI is here to stay and the answers all sound the same. Yes, obviously. No, it is a bubble. Both are answers about the sector, and the sector is not what a business with a phone that rings would be buying.

That business is deciding whether to hand one specific job to a piece of software: answer the phone at seven at night, book the appointment, ask the caller the three questions that show whether the enquiry is worth ringing back. Whether that is a good decision has nothing to do with valuations and everything to do with whether the job is already done well, today, by something you can ring and test.

This post carries no adoption figures and no forecasts, for a reason set out near the end. It carries a method that survives whichever way the sector goes.

What is the difference between a bubble and a tool?

A bubble is a claim about prices. A tool is a claim about a job. The two get run together every time AI is in the news, and running them together is how a business ends up either waiting for a crash that would change nothing about its phones, or buying something because everyone else seems to be. Whether a job is possible is settled by the job, not by the price of whoever sells the means.

So stop asking whether AI is overvalued and ask three narrower questions about the one job you have in mind. Whether it is already possible: you can ring something and hear it done. Whether it is already reliable: its failures are known, bounded and reported rather than silent. Whether it is already cheaper than what you do now, once you count what, if anything, the current arrangement costs you in missed work. A job that passes all three is a tool, whatever the sector is doing. A job that fails one is a bet, and it should be called one.

Which phone jobs are already possible and reliable?

Three phone jobs pass the possible and reliable tests today, and each can be checked on Durbacti's own line rather than taken on our word. The third test, cheaper, is a sum on your own calls, and no post can run it for you.

Answering the phone. Ring 0483 982 033 at any hour and an AI receptionist answers, says it is a virtual receptionist in its first sentence, and takes the call. It answers from the information Durbacti gave it and from nothing else, which is why it does not quote a price nobody set. That is the job worth looking at first, because a call that rings out can go to the next number on the list, and you will not know which ones did.

Booking the appointment. The same receptionist offers a strategy call against a real calendar and books it during the call. A booking is a small transaction with one right answer, which makes it a good job to hand over and a bad one to leave to voicemail.

Qualifying the enquiry. Asking a caller which service they are after, whether they are ready to go ahead and how to reach them is a short list of details it collects on every call. The receptionist asks for the same things on the hundredth call as on the first, and writes the answers into a record a person can act on.

Nothing on that list decides a price, a diagnosis or a dispute. The jobs that pass are the ones a well-briefed new starter could do on their first day with a sheet of answers in front of them.

What does "reliable" mean for a phone call?

Reliable does not mean never wrong. It means the ways the receptionist can be wrong are short, known and visible, so a mistake is found and fixed rather than repeated in silence.

For the receptionist Durbacti builds, that comes down to three limits and a record. It answers only from the information you supplied, so a wrong answer traces back to a fact you gave it and the fix is to change the fact. When the line is bad or the request is unclear it asks a second time instead of guessing. Anything outside its brief goes to a person, under rules agreed before go-live rather than judgement made on the call. And every call leaves a record of what the caller wanted and what happened, which turns "reliable" from a promise into something you can audit on a Friday afternoon.

Those are limits, not features. The receptionist can still mishear a name on a clear line, a fact can go stale, and a rule can be drawn in the wrong place. What happens when the AI receptionist gets it wrong goes through each of those and how they are found. A tool whose failures you can list is one you can run; one whose failures nobody can describe is one to wait on, however confident the sector sounds.

What if the tool you build on disappears?

A tool disappearing is the fair version of the bubble worry. Models get retired. Platforms get bought. Prices change. None of that is a reason to wait, and all of it is a reason to build in a particular way.

Own the record. Every call to Durbacti's own line lands in a database Durbacti controls. A client chooses at scope where their record is delivered: email, SMS, their own CRM or calendar, or the voice platform's dashboard. Choose one you own, and the enquiries and bookings are still yours if the platform vanishes.

Write the brief down. The receptionist's greeting, its facts, its rules and its escalation list exist as a document, not as something that lives inside one vendor's settings screen. A documented brief can be handed to a different platform; one that lives only in a settings screen cannot.

Keep the logic yours and the model swappable. The rules that decide when a call goes to a person are written down before go-live, not learned by a model, and on Durbacti's own line the hours rule is enforced by the database clock rather than by the model. The voice and the transcriber are each one setting, and a model swap is the same kind of change.

What we will not do

•

Predict the sector. We do not know whether AI is overvalued, and neither does anyone selling you a receptionist. Every claim we make is about a job you can test, not a market you cannot.

•

Say AI replaces your staff. An earlier version of this post argued that software wins because people get distracted. That was the wrong argument, and it is withdrawn. The jobs that pass the tests, answering the phone, booking and qualifying, are the ones that ring when nobody on your team is on the phone, and the calls that need a person still go to a person.

•

Quote an adoption figure. How many Australian businesses use AI is a survey result with a date on it, and this post is read years after it was written, so no figure is quoted here. If you want one, read a current survey itself, not a summary of it.

•

Sell a bet as a tool. If the job you have in mind fails one of the three tests, possible, reliable or cheaper, we say so on the strategy call, and the proof of concept is free, so finding out costs you an hour and no money.

What nobody can measure yet

Nobody can measure whether this sector is a bubble. That gets settled by prices over years, and nothing a Sydney business does with its phones changes the answer or depends on it.

Nobody can measure what a job is worth to you until your own calls are counted. The general case is unknowable and the specific one is arithmetic: how many calls you get in a week, what share go unanswered, what share of those were someone who would have booked, and what a job is worth. The missed call calculator puts a ceiling on that sum from those four numbers, all supplied by you, and it is the only figure this post points you at, because it is yours.

What to ask before handing AI a job

•

Can I hear the job done before I pay? A number you can ring, or a working version built on your business. A slide deck is not evidence that it is.

•

What does it answer from? The right answer is "only from what you give it". Anything involving general knowledge can one day say something about your business that you never said.

•

What happens when it cannot help? A list of rules agreed before go-live, not a description of judgement. Ask to see the list.

•

Where does the record go, and who owns it? Your CRM, your calendar, your inbox, a database in your name. A dashboard in the vendor's name is a record you rent.

•

Can the model be swapped without rewriting the rules? If the rules live inside one vendor's model, the vendor is the product.

•

What does it cost to switch off? A monthly fee and a documented brief cost nothing to leave. A long contract and an undocumented brief cost a great deal.

If the job on your mind is the phone, the 24/7 AI phone receptionist page sets out what Durbacti's version does and refuses to do. The free strategy call is where the three tests get applied to your business rather than the sector, and the proof of concept, if the call earns one, is how you hear the answer rather than read it.