We sell one of these, so treat this accordingly. But a list of limits is more useful to you than another list of features, and the questions below are the ones worth putting to anyone selling you AI phone answering — including us.
It can’t quote
Not shouldn’t — can’t, and shouldn’t be allowed to try. Nobody can price a job they haven’t seen, and a number said on the phone becomes the number the customer remembers no matter what gets written down later. That’s how disputes start, and in several states written-quote obligations kick in above certain values anyway.
A call-out fee is different. That’s a fixed, known number and it should be said plainly when asked. Anything past that is “the plumber will confirm”.
It can’t diagnose
It can ask good questions about what’s happening. It cannot tell someone why their hot water has stopped, and it shouldn’t guess, because a wrong guess on the phone sets an expectation the person on site then has to undo.
It can’t talk anyone through a repair
This is the one that matters most. An AI that is helpful by temperament will drift towards being helpful about things it has no business being helpful about. Gas appliances, switchboards, anything involving a ladder or a roof, anything with a tool in the customer’s hand. The instruction has to be a hard stop, not a preference.
The exception is safety, and it runs the other way: get out, turn the mains off at the meter, don’t touch it, ring 000. Those are scripted, checked, and the same every time.
It isn’t a person, and it has to say so
Australia has no AI-specific disclosure law as of 2026, but Australian Consumer Law’s prohibition on misleading conduct doesn’t need one. If a caller asks whether they’re talking to a person, the answer has to be straight and immediate. Anything clever there is a problem waiting to happen, and it’s also just bad manners.
Ask any vendor to demonstrate this. Ring their demo and ask “am I talking to a person?” If the answer is a dodge, that tells you what you need to know.
It’s only as good as the rules you gave it
An AI receptionist doesn’t know your business. It knows what somebody typed into it. If the suburb list is wrong it will book you into a two-hour drive. If nobody told it that you don’t do commercial gas, it will take the booking.
This is why setup matters more than the technology, and why “unlimited AI agents, build your own” is a worse offer than it sounds. Most trade businesses are never going to sit down and configure one properly, and an unconfigured agent is worse than voicemail because it sounds confident while getting it wrong.
It will lose to a person on a strange call
Most calls to a trade business are one of about twenty conversations. The twenty-first — a distressed caller, a complicated commercial arrangement, someone who needs talking down — is where a good human operator is better, and pretending otherwise helps nobody.
What a machine should do there is recognise it’s out of its depth, take a message properly, and get a human to ring back. Ask how a vendor handles that, because “it handles everything” is not an answer.
The questions to ask
- Will it tell a caller it’s an AI if they ask? Ring it and find out.
- What exactly is it forbidden from saying?
- Who writes the rules — you, or them?
- What happens on a call it can’t handle?
- Is the call recorded, is the caller told, and how long is it kept?
If you want to run those questions at ours, the demo takes about a minute to set up.