AI adoption for small business: where to start (and what to skip)

Most of the advice about AI is written for companies with IT departments. Here is what adoption actually looks like when you are a five-person business in Western Sydney and the person reading this is also the person who answers the phone.

The numbers, quickly

AI use in Australian small businesses roughly quadrupled between 2024 and 2025. In accounting firms alone it jumped from 9% to 41% in a single year. So if you have been putting this off, you are not behind some imaginary curve, but your competitors are starting to move. The gap between businesses that use AI and businesses that talk about it is becoming a real commercial gap.

Here is the part the stats miss though. Most of that adoption is shallow. Someone signs up for ChatGPT, asks it to write a Facebook post, gets a mediocre result, and decides AI is overhyped. That is not adoption. That is a free trial that went nowhere.

Why most small businesses get this wrong

I spent twenty years building systems at the big end of town, and the failure pattern there is identical to the one I now see in small businesses. Everyone starts with the tool instead of the problem. They buy the subscription first and go looking for a use second.

Flip it. Forget AI for a moment and answer one question: what task did you do last week that made you think, why am I still doing this manually? For a plumber it is usually quoting. For a cafe owner it is rosters and supplier invoices. For an accountant it is chasing clients for documents. That task, whatever it is for you, is where AI adoption starts. Not with a tool. With a chore you resent.

A sensible first 90 days

Weeks 1 to 2. Track where your admin hours actually go. Not roughly. Actually write it down for ten working days. Most owners guess wrong about their own time, usually by a lot. The thing you think eats your week often is not the thing that eats your week.

Weeks 3 to 4. Take the single worst task from that list and automate just that one. Missed calls going to voicemail? An AI receptionist answers, books the job and texts you a summary. Drowning in receipts? Connect an extractor that reads them and pushes the data into Xero. One task. Done properly. Working every day.

Months 2 to 3. Live with it. Fix what annoys you about it. Once it feels boring and reliable (boring is the goal), add the second automation. Businesses that adopt this way still use their automations a year later. Businesses that roll out five tools in a week usually abandon all five.

What to skip

Skip anything pitched as an all-in-one AI business platform. Skip tools that need you to change how your whole business runs before they deliver value. And skip writing your own prompts from scratch for repetitive work. A prompt library tuned to your business, your tone and your documents will outperform improvised prompting every single time, and it takes an afternoon to set up.

One more thing worth skipping: the guilt. You do not need to understand how a language model works to benefit from one, the same way you do not need to understand a gearbox to drive a ute.

The honest cost picture

A useful single automation runs a few hundred dollars a month. Set against three to five recovered hours a week, it pays for itself in the first fortnight, and that is before counting the jobs you stop losing to missed calls. If a provider cannot show you that maths for your specific business, keep your wallet in your pocket.