Tradies expect an AI receptionist to mean a changed number, new software and a week off the tools. None of that is true. Here is how Durbacti connects a receptionist to a plumber's or sparkie's existing number through Twilio and Vapi, what it asks them to have ready, the one list that cannot be skipped, and the part that takes an hour rather than a week.
A buyer asking about an open time can wait until morning. A tenant with no hot water or a door that will not lock cannot, and the law treats the two differently. Here are the rules Durbacti would write for a tenant's after-hours repair call, where they come from, and the two decisions the receptionist must never make alone.
A venue buys an AI receptionist for the Friday rush and the 10 pm function enquiry. Here are the six after-hours calls it should never finish, from an emergency to a chef who cannot make the lunch shift, and what it does instead.
Every page here that describes the receptionist says it answers from your information, and none shows what that looks like written down. Here is the shape of Durbacti's own brief, paragraph by paragraph, what a restaurant's would say, and seven questions to answer before writing yours.
The site says the receptionist transfers "under rules you set" and never gives the list. Here it is: the seven lines Durbacti's own receptionist works to, which of them hand a call to a person, the one a clock enforces rather than the model, and how to write yours.
Vendors say "integrates with your PMS" and mean one of three things. Which practice management systems let a third party write an appointment, why the receptionist inherits your online booking rules, and when a message is the right answer.
An AI assistant can read your website. Using it is different. What WebMCP is, what we described on durbacti.com, and why describing your forms costs almost nothing even if WebMCP never ships.
Ring a demo line with the vendor's own questions and it proves the vendor's script. Nine calls that test an AI receptionist on your calls instead, what to listen for on each, and why two fails matter more than the other seven.
The missed call figures in one web search, all from companies selling call answering, describe different businesses, and none of them is yours. How to test any of them, and how to measure your own missed calls from one ordinary week of call records.
The question buyers ask quietly, after the demo: what happens when it answers confidently and is wrong? An error is a different failure from an escalation. Here is what a wrong answer can and cannot be, how you find out, and who fixes it.
It is the question that comes up right after "does it actually sound human": are we allowed to record these calls? Here is the shape of the problem, the practical answer with the least to argue about, and the specific questions worth taking to a lawyer before you switch anything on.
Search used to end with a list of links. Increasingly it ends with an answer, and the answer either mentions you or it does not. Here is exactly what we changed on this site to make it legible to AI assistants, what we deliberately refused to do, and what we still cannot measure.
When the Stargate project was announced in January 2025, a $500 billion bet on AI infrastructure sounded abstract. By mid-2026 its effects were showing up in the tools small businesses use every day. Here is what actually changed, and what it means for Sydney businesses.
The moment to answer a buyer is while they are still on the page, and a reply the next morning can land after they have moved on. What a chatbot changes at each stage of a B2B funnel, what it cannot do, and what to measure before deciding.
Whether AI is a bubble is the wrong question for a business with a phone that rings. The useful one is whether the specific job you would hand it is already possible and reliable, and whether it is cheaper on your own numbers.