Yes — a small business can put AI to work on its admin, its customer replies, its bookkeeping prep and its marketing without hiring anyone, and in 2026 the owner who does it is the norm in the businesses that are growing. What you cannot do without help is build custom software, connect systems that do not want to be connected, or hand a model decisions that cost real money when they are wrong. The line between the two is the subject of this piece, along with the order to do things in and the mistakes that cost owners the most.
We are an Australian company and the examples are Australian; the approach is not.
What can a small business realistically do with AI on its own?
Four things, and each of them pays for itself in the first month:
- Customer replies. Enquiry emails, quote requests, the same eleven questions every week. A model drafts the reply from your notes and your price list; you read it and send. Most owners get an hour a day back here.
- Admin and paperwork. Meeting notes into actions, supplier emails into a summary, a policy or a contract into plain English, the tender question you have answered eight times answered a ninth. The AI for Operations course is built around this list.
- Marketing. First drafts of everything — posts, emails, the website copy you have been meaning to fix — from a brief you write once. Should a marketer learn to code to use AI covers the method; the answer for owners is the same: the brief is the skill.
- One automation. New leads from the website into a spreadsheet with a draft reply; invoices reconciled against the bank feed; a weekly digest of what changed. The no-code tools (n8n, Make, Zapier) do this now without a developer.
What ties these together is that in each case the model produces a draft and a person sends it. That is the boundary that keeps a small business safe.
What should a small business not try to do alone?
Anything where the model acts without a person checking, on something that costs money or trust when it is wrong. Concretely: an AI that answers customers without review, an automation that moves money, anything that touches payroll, and any decision about a person — hiring, credit, a complaint — where a wrong answer is a legal problem. These are not impossible; they are second-year projects that need someone who understands failure modes, and the cost of getting one wrong once is higher than the cost of a contractor.
The other thing to leave alone is custom software. If the need is "we want our own app", that is a developer's job — though it is now a much smaller job than it was, and a developer with a coding agent can build in a week what took a quarter. Ask for a fixed-price quote.
Which tools, and what do they cost?
A frontier-model subscription for the owner and anyone who writes — A$30–A$40 a month each — and one automation tool, A$0–A$50 a month. That is the whole stack for the first six months. Choose the model whose company you are comfortable giving your business documents to, and read the data terms: the free consumer tiers of most tools may use what you type for training, the paid business tiers generally do not, and the difference matters the moment a customer's details are in the conversation.
Do not buy "an AI solution for small business" from anyone in the first six months. Nearly all of them are the same model behind a form, at ten times the price, and you will not know whether it is any good until you have used the underlying tool yourself.
What about our customers' data and the Privacy Act?
Treat every AI tool as a third party you are sharing data with, because that is what it is. Australian small businesses under A$3 million in turnover are mostly exempt from the Privacy Act, but the exemption is under review, it does not cover health providers or businesses that trade in personal information, and customers do not care about the exemption when their details leak. The practical rules:
- Customer names, contact details and anything about their circumstances go only into a paid business tier whose terms say it does not train on your data.
- Financial records and anything under a confidentiality clause with a supplier or client: same rule, and check the clause.
- Write down which tools are approved and for what, on one page, and show it to anyone who joins. How to write an AI usage policy for your workplace has a template that takes an afternoon.
In what order should an owner do this?
Replies first, then admin, then marketing, then the automation — because that is the order of payback, and because the first three teach you what the model gets wrong before you let it run unattended in the fourth. Give it a month per step. By month four you have a business where the routine correspondence drafts itself, the paperwork is summarised before you read it, the marketing is going out on schedule, and one recurring task runs on its own.
The mistake owners make is starting with the automation, because it is the impressive one. It is also the one that fails silently at 2 a.m. and sends forty customers the wrong thing.
When is it time to hire or bring in help?
When the automations you want are the kind that act rather than draft, when you have three systems that need to talk to each other, or when the owner is spending more than a day a week on the tooling. At that point a contractor for a week — or a part-time AI automation engineer, a role that did not exist three years ago — is the right spend, and by then you will know exactly what to ask for, because you have done the first year yourself.
The AI for Founders course is the owner's version of this whole piece with the work graded; the for teams page is for the business that has grown past one person doing it. The free AI for Founders skill check takes three minutes with a free student account and tells you which of the four steps you are on.