The AI Skills Gap Nobody's Talking About: Why New Zealand Small Businesses Are Using AI for the Wrong Things

You Are Using AI for the Wrong Thing

Captions are the lowest-value use available to you · Updated July 2026

The tradie problem

There is a specific kind of business owner New Zealand is full of. They are excellent at the actual work. Twenty years on the tools. Their customers love them. And they are quietly bleeding money in places they cannot see, because seeing them requires expertise they were never trained in.

Call it the tradie problem, though it applies just as much to a cafe owner, a boutique retailer, or a plumber with four vans.

The blind spots are almost always the same, and in New Zealand there are four rather than three.

Revenue and enquiry. They know how many jobs they did last month. They do not know how many enquiries they received, what percentage converted, or which lead source produced the highest-value work.

Legal. They sign contracts they have not read properly. Their terms of trade were copied off a competitor's site in 2019.

Accounting. They look at the bank balance and call it a financial system. Their accountant does compliance, not analysis.

Health and safety. Under HSWA every PCBU has duties, and being a small operator is not a defence. Most sole traders and small crews have documentation that would not survive scrutiny.

And here is what most of these operators are doing with AI: generating Instagram captions.

That is the whole thing in one sentence.

Why the caption thing happened

It is not stupidity. It is discoverability. Content generation is the most visible, most demoed, most heavily marketed AI use case, so it reached small business owners first. It also feels safe. If the caption is bad, you delete it.

The higher-value uses feel riskier because they touch things that matter. Asking AI to help you understand a contract feels like it should require a lawyer. Asking it to analyse your enquiry data feels like it should require an analyst. So people default to the low-stakes toy.

But the risk calculation is backwards. The alternative to using AI on that contract is not "get a lawyer to read it". The alternative is not reading it properly at all. That is the actual comparison. Not AI versus expert. AI versus nothing.

Trade businesses: the enquiry leak

Take a typical electrical business. Three vans, owner-operator, roughly 40 enquiries a month across phone, website form and word of mouth. He quotes maybe 25 and wins maybe 10.

He experiences this as "we're pretty busy". What he cannot see is that 15 enquiries a month never got quoted, some because he was on a job and did not ring back for two days, some because he sized them up as tyre-kickers on the phone. He has never checked whether that instinct is accurate.

The AI use here is not writing a quote. It is this: export twelve months of enquiries into a spreadsheet, feed it in, and ask what patterns exist. Which source produces the highest average job value. What the relationship is between response time and conversion. Whether the jobs he declined to quote actually resemble the jobs he later won.

That is a two-hour exercise most operators would never commission, because a consultant would charge thousands for it. The answer is usually confronting and immediately actionable: your Tuesday enquiries convert at half the rate of your Thursday enquiries, because Tuesday is your busiest install day and you do not ring back.

The second trade-specific use is health and safety documentation. HSWA obligations are real and the paperwork burden falls hardest on the operators least equipped to handle it. AI will not write your safety system for you, but it will help you read WorkSafe's guidance, work out which of your practices are documented and which are just habit, and produce a first draft of a task analysis you can then have reviewed properly.

[INTERNAL LINK 1 — tracking where enquiries actually come from]

Cafes and restaurants: the menu margin problem

Hospitality operators are famously good at food and famously bad at unit economics. The classic scenario is a cafe owner with a menu of 30 items who has never costed more than about six of them properly.

Here is a use case that takes an afternoon. Photograph your supplier invoices for a month. Photograph the menu. Ask AI to build a per-item cost estimate against sell price, flagging where the gap is thinnest.

Work in GST-exclusive figures throughout. At 15 per cent, mixing inclusive and exclusive numbers will distort every margin you calculate, and that single error is more common than you would think.

It will be imprecise. Wastage, prep labour and portion drift all mess with it. But imprecise and directional beats absent. Most operators discover the same thing: their most popular item is one of their least profitable, and it is popular partly because it is underpriced.

The second use is better. Point AI at your POS export and ask what sells together and at what time. Not to write a marketing campaign, but to change what you prep and when.

Small retail: stock and the enquiry gap

The independent retailer's version is stock. Which lines tie up capital for months. Which supplier's terms are quietly costing more than their prices suggest. Which products get enquired about and never bought, which is the single most useful and least-tracked signal in retail.

There is also a legal use case retailers specifically need. Consumer guarantees under the Consumer Guarantees Act are non-negotiable, and the Fair Trading Act sets rules about how you represent your products and your refund policy. Both are enforced by the Commerce Commission.

Plenty of small retailers have refund signage that misrepresents customers' actual rights, with "no refunds on sale items" the classic example. That is not merely unenforceable, it is potentially a Fair Trading Act breach in itself. Reading your own policy against the Commission's guidance is a 20-minute exercise with AI as a reading partner, and it is worth vastly more than a month of Instagram captions.

The safeguarding use case

This is the one nobody markets, because it does not produce a shareable output.

Contracts. Every trade business signs subcontractor agreements, supplier terms and commercial leases. Almost none are read line by line. Feeding a contract in and asking what obligations it creates, what happens if you want out, and what is unusual compared to standard terms for that type of agreement is not legal advice. But it tells you which three clauses to actually pay a lawyer to look at. It converts "I can't afford a lawyer" into "I can afford twenty minutes of a lawyer".

Same logic on the accounting side. You are not replacing your accountant, you are arriving at the meeting with questions instead of a shoebox. And if you are anywhere near the $60,000 GST registration threshold, that is a conversation worth having before you cross it rather than after. business.govt.nz is a reasonable reference to have open while you work through any of it.

Privacy: where New Zealand differs from Australia

This one deserves its own section, because a lot of Kiwi operators have absorbed Australian advice that does not apply to them.

The Privacy Act 2020 applies to every agency handling personal information. There is no small business exemption. Australia has one, which is why so much of the content circulating online tells small operators they can relax, and why plenty of New Zealand businesses believe they are outside the rules when they are not.

If you run a booking system, a loyalty programme, a customer database or an email list, you have obligations under the information privacy principles, covering collection, storage, access and correction, plus the mandatory notifiable privacy breach scheme.

Most small operators have never read the principles. Reading them against your actual practices, with AI helping you translate legislative language into "does this apply to my POS system", is genuinely one of the highest-value hours you could spend. Start with the Privacy Commissioner's guidance on the information privacy principles.

[INTERNAL LINK 2 — the admin side of running a business]

Start with the question you have been avoiding

Here is the practical way in, and it takes about a minute.

Think of the thing in your business you have been putting off because you do not know how to think about it. Not the thing you have not got round to, the thing you genuinely do not know how to approach. The lease renewal. Whether your best-selling job type is actually your most profitable. Whether your safety documentation would hold up.

That discomfort is the signal. Owners avoid those questions precisely because answering them has always required expertise they do not have and cannot cheaply buy, so the question gets filed under "deal with it later" indefinitely. It is the highest-value question in your business by definition, because it is the one that has gone unexamined longest.

Start there. Not with the thing you already know how to do faster.

What this means for how you spend your twenty minutes

The honest framing: AI has not made small business owners better marketers. It has made them faster at producing mediocre marketing. That is a real but modest gain.

What it has genuinely done is give a solo operator access to a category of thinking that was previously locked behind professional fees. Analysis. Interpretation. A second opinion on something you do not have the training to evaluate alone.

So if you are going to spend twenty minutes on AI this week, do not spend it on a caption. Spend it asking a hard question about a part of your business you have been avoiding.

[INTERNAL LINK 3 — using AI properly in your business]

Frequently asked questions