AI voice agents

What should an AI receptionist do for a plumber, electrician or other trade?

By the Cognexa team · 30 October 2026

It should take the call you can't, fill in a proper job card, and get the right calls to you while you're on site. That means asking where the job is, what's wrong and how urgent it is, checking the address against the areas you cover, and booking a site visit into the diary of whoever is available. The calls you've marked urgent ring your phone straight away. Everything else becomes a booking or a callback with the details already written down, so you don't have to ring a customer back to ask what they wanted.

You know the moment. You're in a roof, or halfway through a DB board, and the phone goes. Getting the call answered at all is the easy part, and we've covered the phone settings, callbacks and after-hours cover in how to stop missing customer calls. This post is about what happens after the assistant says hello, and what it writes down for you.

The job card it should fill in on every call

A message that says "Mrs Naidoo called about a leak" is nearly useless. You still have to phone her, find out where she lives, and work out whether it's tonight's problem or next week's.

An AI receptionist can qualify a lead by asking what, where and when. For a trade, that becomes a short job card:

the street address and suburb, checked against the suburbs you actually cover
what's wrong, in the caller's own words ("the geyser's dripping through the ceiling", "the kitchen plugs have all died")
whether it's happening right now, or something they've noticed and want looked at
how you get in: a complex gate, a dog in the yard, a tenant who has to be home
whether the caller owns the property or rents it, if your business needs the owner's go-ahead before you start
the best number to reach them, confirmed back to them

The suburb check matters more than it looks. A caller who lives just outside your area should hear that on the call, politely, before they wait for a callback that won't come. The assistant can only do that if your knowledge base lists the suburbs you serve and the work you don't do, which is part of how we build every AI receptionist.

Keep the card to what the job needs. POPIA's section on minimality says personal information may only be processed if it's adequate, relevant and not excessive for the purpose. Booking a plumber's visit doesn't need the caller's ID number, so the card shouldn't ask for it. The exact wording and order of the questions belong in your script, and we've set that out, with a plumber as the worked example, in what your AI receptionist script should say.

Decide which calls ring your phone on site

Write this rule down first: the assistant needs it most.

Make a short list of what's worth interrupting a job for. Water coming through a ceiling. No power to the whole house. A customer you're already working for who says something has gone wrong since you left. A caller who asks for you by name. Those transfer live, and if you can't pick up, the assistant books a callback and tells the caller so. Our receptionist page sets that rule out: it transfers where somebody is available and books a callback where nobody is.

Everything else waits for you. A quote for new downlights, a question about whether you do solar, a supplier chasing payment. Cold sales calls and spam get screened before they reach you at all.

The assistant shouldn't judge whether something is dangerous. If a caller describes anything that sounds unsafe, the rule is simple: put them through to a person. And who gets the next emergency slot, or which job gets bumped, stays your call. Our workflow automation page is plain about that split: software books the routine work, and a person decides emergency slots and who gets priority.

Getting a quote request into the diary

When a call is a quote request and the job needs to be seen, the useful outcome is a site visit with a time on it.

To book that, the assistant needs to see real availability. Google's Calendar API can report which times are busy across a set of calendars, so the assistant can look across each technician's diary before it offers a slot. That's the build our pricing page describes at the top of the range: a voice agent that qualifies the caller, checks a technician's availability, books the job, raises the quote and logs it against a CRM record.

What the assistant may say about price is up to you. If you give it a call-out rate or a price list, it can quote from that. If you don't, it says a technician will quote on site and books the visit. Where a job can be priced from a description, the quoting side of our workflow automation asks the same questions on every channel, then your rate card, travel bands and VAT do the arithmetic. Special pricing, odd jobs and whether you want the work at all stay with you.

Photos, and the WhatsApp rule to plan for

A photo of the problem can save a wasted trip. A picture of the DB board or the geyser's pressure valve can tell you which parts to load before you leave.

WhatsApp is the natural way to get it. When a customer sends an image to a business on the WhatsApp Business Platform, it reaches the business's connected system, so the photo can sit with the job card.

Here's the catch. Meta only lets a business send ordinary WhatsApp messages inside a customer service window, which opens when the customer messages or calls you on WhatsApp. A phone call to your normal business number doesn't open it. So the "please send us a photo" message after a call has to be a pre-approved template, and the wording needs approval before the assistant goes live. The assistant asks on the call whether the caller is happy to get a WhatsApp from you, because Meta only allows messages to people who have opted in. A compliant build gets that wording written and approved up front. If you'd rather keep it simpler, the assistant can give the caller your WhatsApp number on the call. When they message you with the photo, the window opens and your reply can be an ordinary message.

How we set it up for a trade business

We start by scoping what the assistant must handle and what it must never touch. Then we write the knowledge base from your own material: your services, trading hours, the suburbs you cover and the work you don't do.

Next we connect the systems. That means the diary, and a CRM, a quoting tool or WhatsApp if you use them. Before any customer reaches it, we test it against awkward callers. It goes live on a narrow scope and widens as you trust it. If you want calls to reach it only when you don't answer, call forwarding keeps your existing number.

What it costs depends on how many calls you get and how many systems it reads from and writes into. Answering and taking a job card is a smaller build than checking technicians' diaries and raising quotes. You see the quote before you commit, and the pricing page explains what moves it.

Before you speak to anyone, write down your urgent-call list and the suburbs you cover. Then tell us what your phone does on a bad day, and we'll map which of those calls an assistant should take off you.