A geyser bursts in a townhouse in Durbanville at 19:40 on a Tuesday. The owner does what everyone does now: opens WhatsApp, finds three plumbers a neighbour once recommended, and messages all three. Whoever answers first, with a sensible question and a rough idea of when they can come, gets the job. The other two reply at 07:30 the next morning, by which point there is water in the ceiling and a competitor's van in the driveway.

This is the actual shape of the problem for most South African plumbers and electricians. It is not that the work is hard to find. It is that the work finds you at the worst possible moment — mid-job, hands full, on a roof, under a sink — and the customer's patience is measured in minutes.

The trades are where I see the widest gap between what AI is sold as and what actually helps. Nobody quoting a distribution-board upgrade in Mitchells Plain needs a chatbot that pretends to be a person. What they need is narrower, and far more useful: to never miss the after-hours call-out, to get a quote out before the customer cools off, and to stop spending Sunday evening typing the same job updates one thumb at a time.

The after-hours lead you are quietly losing

Most one-van and small trade businesses have no office. The phone is the owner's personal cell, and it rings while he is wrist-deep in a job he is already being paid for. He cannot answer. He tells himself he will call back at five. Sometimes he does. Often, by five, the customer has booked someone else.

You can measure this loss, roughly. If you miss two genuine call-outs a week, and one in three of those would have converted at an average job value of R2 800, that is a number worth saying out loud: somewhere around R8 000 a month walking past your door because nobody picked up. Most trade owners have never put a figure on it because the lost jobs never appear anywhere — there is no record of the call you didn't take.

An AI answering layer changes the economics here without changing how you work. When a call comes in and you don't pick up, it doesn't go to voicemail (which almost nobody in South Africa leaves any more). Instead the caller gets a WhatsApp within seconds: a short, plainly-worded message that asks the three things you'd ask anyway — what's the problem, where are you, and is it an emergency. The customer answers in their own time, the details land in one place, and you call back already knowing whether you're driving out tonight or quoting on Thursday.

The point is not that AI handles the job. It is that AI holds the lead still long enough for you to handle it.

Triage that knows the difference between a drip and a flood

Not every after-hours message is an emergency, and treating them all as one is how trade owners burn out. A blocked drain that can wait until morning should not pull you off the couch. A live electrical fault in a house with small children should not wait in a queue behind a quote request for a new plug point.

This is where a well-built intake flow earns its place. The system asks a couple of follow-up questions and sorts the message into rough buckets — emergency, book-soon, quote-only — based on what the customer describes. Water coming through a ceiling, no power to half the house, gas smell: these get flagged and pushed to you immediately, with a notification you can't ignore. A request to quote on recessed lighting in a renovation gets logged for normal hours.

A caution here, because it matters. The triage must be conservative. It should be biased towards escalating, not towards deciding something is routine. If the customer's description is ambiguous, the system treats it as urgent and gets it in front of a human. An AI that confidently downgrades a real emergency to "we'll get to it tomorrow" is worse than no system at all. I would rather a plumber get three false alarms a month than miss one real one — and any setup that gets this backwards should be thrown out.

Quoting faster without quoting badly

The quote is where trade businesses leak the most money, and not in the way people assume. It is rarely the price that loses the job. It is the delay. The customer who waited four days for your quote has, in those four days, received two others and emotionally committed to whoever felt most on-the-ball.

Quoting is also genuinely hard to automate well, and I want to be honest about that. A plumber pricing a re-pipe is drawing on twenty years of knowing what's likely behind that wall. No AI is going to replace that judgement, and you shouldn't let one try.

What AI can do is take the judgement you've already applied and turn it into a sent quote in minutes instead of days. Here is the realistic version of the workflow:

The customer gets a clean, professional quote the same afternoon. You never sat down at a laptop. And critically, the figures are yours — the AI assembled the document, it did not decide the price.

For electricians, there's an added layer worth building in. A quote for installation work that requires a Certificate of Compliance can automatically carry the right language about the CoC, what it covers, and that it will be issued on completion in line with the wiring code. It keeps you consistent on the compliance wording without you having to remember to paste it in every time.

Job-site updates the customer actually reads

On a multi-day job — a bathroom re-do, a board upgrade, a solar-ready DB change — the customer's anxiety isn't usually about the work. It's about silence. They don't know if the part arrived, whether you're coming Thursday or Friday, why there was no-one on site this morning. So they phone. And you stop work to reassure them.

A light-touch update flow handles the routine end of this. When you mark a job stage done — "first fix complete", "waiting on the geyser from the supplier", "testing tomorrow morning" — the customer gets a short, human-sounding WhatsApp in plain language. Not a barrage. One clear message at the moments that matter. When the supplier is the hold-up, the customer knows it's the supplier and not you, which protects the relationship in exactly the spot where trade reputations get damaged.

One thing AI should not do here is invent detail. The update can only say what you've actually told it. A system that generates cheerful, plausible-sounding progress reports it has no basis for is a liability — it will eventually tell a customer the job is on track on a day you didn't go in. Keep the updates tied to real status that you set.

Where this fits with the tools you already pay for

Most SA trade businesses are already half-digitised, often without realising it. You might be invoicing through Xero or Sage, taking card payments on a Yoco or iKhokha machine, keeping customer numbers in your phone and job notes in a WhatsApp chat with yourself. The automation worth building sits on top of what you've got, rather than asking you to move to some new platform you'll abandon in a month.

The honest version of this is unglamorous. A new lead becomes a customer record. A confirmed job becomes a calendar entry and, when done, a draft invoice in the accounting tool you already use. The WhatsApp number you already give out becomes the front door, with the AI sitting behind it to catch what you can't. Nobody learns a new app. The plumber still plumbs.

I'd steer most trade owners away from two things, at least to begin with. Voice AI that answers and holds a full conversation with the caller — the tech works, but South African customers calling a tradesman want to reach a person, and a robot voice on a burst-geyser call reads as a business too big to care. And anything that auto-sends quotes or prices without you in the loop. The whole value of a trade business is that a human who knows the work stands behind the number. Automate the typing, the chasing, the catching. Not the judgement.

A sensible first project

If you run a plumbing or electrical business and want one thing that pays for itself quickly, start with after-hours lead capture and triage. It is the cheapest to build, the easiest to measure, and it addresses the loss that hurts most — the job that went to the competitor because you were under a sink when they called.

Run it for a month and the numbers tell you whether to go further. Count the leads it catches outside working hours that you'd otherwise have lost. If that number covers the cost several times over — and for most working trade businesses it does within the first few weeks — then layer on the quote-drafting next, and job updates after that. Build it in the order of what's actually costing you, not in the order of what sounds most impressive.

The trades don't need reinventing. The work is the work, and the good ones are already good at it. What a sensibly-built bit of automation does is stop the phone, the quote, and the silence from quietly costing you customers you'd already half-won. That's a narrow promise. It's also one I've watched hold up on real jobs, in real Cape Town suburbs, for owners who never wanted to become tech companies and shouldn't have to.