Most "AI for tradies" lists are affiliate pages. This one names what doesn’t work.

What’s actually useful on a job site

A short list, and it is short because the useful cases share one property: the input is unstructured and a human currently reads it.

That is the entire test. If a person is currently reading something and deciding, an AI step might help. If a person is currently following a rule, a rule will do it cheaper and more reliably.

• Reading supplier invoices They arrive in a different format every month, which is exactly why rules fail and why this is the strongest case on the list.
• Triaging an inbox Sorting genuine enquiries from suppliers, spam and internal traffic. Judgement, low stakes, high volume — a good fit.
• Taking a call at 11pm Capturing enough detail that someone can call back in the morning. Not closing the job; capturing it.
• Turning a voice note into a job record You talk on the drive home, it becomes structured. Genuinely useful because the alternative is it never gets written down.

What’s a novelty

Anything that produces text you then have to check line by line. If verifying the output takes as long as writing it, the tool has moved the work rather than removed it.

Chatbots on trade websites are the clearest example. Your customers ring. A chatbot intercepts the small fraction who would have filled in a form anyway, and irritates some of the rest. We decline this work — it is on the list of what we do not do.

Also novelty: AI-generated quotes. The number is the one thing on a quote that must be right, and a model that is right most of the time is worse than a template that is right every time.

AI that costs more than the problem

A model call has a unit cost. A rule does not. So anything with a fixed, knowable answer belongs in a rule, and putting it in a model means paying every month for a decision that was never in doubt.

The pattern we see most: an AI step doing classification that three if-statements would handle. It is more interesting to build, it demos better, and it bills forever. We move work out of the AI step and into rules wherever we can, and that reduces what you pay us monthly — which is the correct incentive and an unusual one.

Low-frequency work is the other trap. Something you do four times a year does not justify a build at all, however good the tool — the full list of what not to automate covers why.

Quoting, invoicing and paperwork: where it genuinely helps

Not in writing the quote. In everything around it: reading what came in, getting it into the job system without re-typing, and chasing what has not come back.

That distinction matters because it is where the money actually is. The quote takes twenty minutes and someone is being paid to think during it. The admin around the quote takes longer, produces nothing, and is where the boring jobs argument comes from.

Most of what looks like an AI win in this area is actually an integration win — the information moving without a human retyping it. That does not need a model at all.

What we won’t claim on this page

We have not surveyed tradies, and we are not going to write "what tradies tell us doesn’t work" as though we had. Kindra AI was founded in 2025 and the sample is not large enough to publish patterns from.

What is on this page is the view from building the systems: which cases pay, which bill forever for nothing, and which we decline. That is a narrower claim than most pages of this kind make, and it is the one we can actually support.

When there is enough measured client data to say what fails in the field, it will be on this page with the sample size attached. Until then, treat the sections above as a builder’s opinion rather than research — an honest label, and a rarer one than it should be.

Where a built system beats a tool

When the AI step needs to sit inside your workflow rather than beside it. A tool that reads invoices and emails you a summary has not saved you much; the same reading, posting straight into your accounting package, has.

That is the difference between buying a product and building the joinery, and it is why most of what we do is the connections rather than the intelligence. The model is often the cheapest part.

If you want to know whether your business is at that point, the readiness scorecard takes three minutes and has no email gate on it.