Under the hood
What the assistant actually does
'AI chatbot' covers everything from a scripted menu to a model left free to invent, so here is the specific shape of what we build. The assistant is trained on your own material — services, inclusions, service areas, policies, hours and the answers you already give on the phone. It answers from that, and when it does not have an answer it says so and moves to collecting the enquiry instead of improvising one.
Underneath the conversation sits a qualifying path. You decide what a real enquiry looks like — the location, the job type, the budget band, the timeframe — and the assistant asks for those in the flow of a conversation rather than confronting a first-time visitor with a nine-field form. Location is the one that has to be set up for your market: a suburb, a postcode, a PIN code or a ZIP code, in whatever form your customers would type it.
When someone qualifies, the details go where you already work: an email or SMS alert, your CRM, your shared inbox, or a booking slot if you take appointments. The transcript travels with the lead, so you open a job already knowing what was asked and what was promised.
Hand-off rules are part of the build rather than an afterthought. Anything sensitive, urgent or outside the trained material is escalated to a person, and after-hours conversations are captured with a clear statement about when a human will reply. An assistant that quietly pretends to be you is worse than no assistant.
After launch we read the real conversations. The ones it answered badly become new trained answers; the ones where it handed over too early tighten the rules. That tuning stage is why the first few weeks of live traffic teach us more than the training material ever does.