The 67% problem — why most Aussie SMBs still can't use AI

Every survey run on Australian small business and AI adoption over the last three years has landed on roughly the same number. Two-thirds want it. Two-thirds have tried it. Two-thirds don't currently use it. The problem isn't awareness or willingness — the problem is that the first attempt didn't stick.

Here's what the 67% actually experienced.

01The five reasons the first attempt fails

1 · The tool was general when they needed specific

The most common first-experience is buying a general-purpose AI assistant (a ChatGPT Plus, a Microsoft Copilot, a bought SaaS with an 'AI features' badge). The owner asks it to do a specific job — chase this invoice, draft this quote, follow up this client — and it produces something plausible but wrong on the specifics. The client's first name is wrong. The context is generic. The tone isn't theirs. It sounds like an American temp wrote it.

That's not a bad AI. It's an AI that hasn't been given the context to be good at the specific task. But the owner doesn't know that — they just feel burned.

2 · The tool didn't read the business data

The value in AI for a small business is in reading your data — Xero, CRM, inbox — and producing output that uses the real numbers, the real names, the real history. Almost every off-the-shelf AI tool doesn't touch business data. It reads what you paste in. That's fine for content but useless for operations.

3 · Nobody set it up properly

A general-purpose AI needs a bit of setup to be useful. It needs your voice examples pasted in. It needs the exclusions ('don't invent numbers, don't add scope not requested'). It needs the format you want the output in. Most owners never do this — they type a bare prompt, get a generic answer, and conclude AI isn't for them. It's not the AI; the tool needed 20 minutes of setup and got zero.

4 · The task wasn't worth automating

This is the quiet one. The task was small enough that doing it manually took 15 minutes, and automating it (learning the tool, prompting it, editing the output) took 20. Owners get sold on AI-for-everything and end up automating things that were fine manually. The energy drain of learning a new tool for no net win kills the willingness to try the next thing.

5 · The organisation isn't ready

The business has three people, five spreadsheets, and no shared documentation on how anything works. AI can't function well in that environment because there's no organised context to feed it. Sixty percent of the AI-fit problem for small business is really an organisation problem in disguise.

02What actually works

The Brisbane SMBs we've seen get real value from AI have all done the same four things:

• Picked one specific task that takes them >2 hours a week manually.
• Fed the AI real business data — voice examples, past invoices, client history.
• Accepted that the first two weeks look worse than the manual process.
• Refined it in weeks 3–6 until the AI output was 90%+ approve-and-send.

That's the whole trick. Nothing exotic. Owners who skip step 3 (the awkward learning weeks) cancel and become the 67%.

03The blind spot

There's a specific blind spot in the 67% number. Most surveys ask about using AI — not about having AI used on the business. Those are different things. A custom build (someone else does the setup, the training, the refinement, then hands you the working tool) skips every failure mode listed above. The owner never has to learn a prompt or paste examples. The tool just works.

This is why we build custom apps for small business instead of selling AI training courses. The 67% isn't fixable one owner at a time. It's fixable by taking the setup pain off their plate entirely.

04The realistic path forward

If you're in the 67%, here's the two-step:

• Step 1: Pick the single most-annoying repetitive job in your week. Something you do >2 hours on. Write it down in one page of plain English — what happens, why it takes so long, what the ideal outcome looks like.
• Step 2: Get a Leak Check. We'll tell you honestly whether it's worth building a custom tool for that job, or whether thirty minutes of setup on a $30/month SaaS would do the same thing.

The honest answer is often the second one — plenty of tasks are better served by cheap SaaS with a bit of setup, and I'll say that clearly. Custom's for the specific-to-you jobs where the SaaS setup can't quite fit.