AI adoption in Australian small business: why the numbers disagree
One source says 12% of Australian businesses use AI. Another says 72% of small businesses have. Here is what each figure really measures and what it means for you.
Read articleAutomation
AI agents can draft, sort and follow up on routine work. Here is what they are, which jobs suit them, and where a person should always stay in control.

Quick answer: AI agents for small business work best on narrow, repeatable jobs where a person can easily check the result. Start with one low-risk task, let the agent draft or sort the work, and keep a person approving anything that affects money, promises or customers.
A chatbot answers when you ask. An AI agent goes further: it is given a goal and can take several steps, often using other tools such as your email, calendar or customer records.
An agent could read a new website enquiry, work out what the person needs, draft a reply and suggest who should handle it. Ordinary automation follows fixed rules. An agent can deal with messier inputs, but it can also get things wrong in less predictable ways.
A good first job has four qualities. It happens often, it follows a similar pattern each time, mistakes are cheap to fix, and a person can check the result quickly.
| Job | What the agent can do | What a person does |
|---|---|---|
| New enquiries | Sort, summarise and draft a reply | Reviews and sends |
| Meeting notes | Summarise and list actions | Corrects and shares |
| Marketing content | Draft variations from approved material | Edits and publishes |
| Reporting | Gather figures into a summary | Checks numbers and draws conclusions |
| Data clean-up | Flag duplicates and missing details | Approves changes |
The National AI Centre’s guidance asks organisations to match oversight to how autonomous a system is and how high the stakes are. Automated monitoring may be fine for low-stakes uses, while high-stakes decisions call for mandatory human review. It also recommends clear points where people can pause, roll back or shut a system down.
In practice, keep a person approving anything that:
Write down the steps as they happen today, including who does them and what information they need. If you cannot describe the process, an agent will not fix it.
Give the agent access only to what the job needs. Read-only access is a sensible way to begin.
Have the agent prepare work without sending or changing anything. Compare its drafts against what your team would have done.
Use incomplete enquiries, duplicates and unusual requests. Decide how the agent should hand over when it is unsure.
Once results are reliable, allow it to complete the lowest-risk steps by itself. Keep reviewing a sample of its work.
Our short guide to choosing your first marketing automation follows the same principle, and our marketing automation service can help you set it up.
The OAIC’s guidance on commercially available AI products is practical reading. It recommends not entering personal information, especially sensitive information, into publicly available generative AI tools. It also says a person should verify the accuracy of AI-obtained personal information and be able to overturn AI decisions.
Clearly identify public-facing AI tools, such as chatbots, so people know they are not talking to a human. Update your privacy policy and collection notices to explain your use of AI.
If an agent helps with decisions that could significantly affect people, read our guide to the automated decision making privacy rule that begins in December 2026.
This area attracts bold claims. While researching this article, we found widely repeated productivity statistics with no traceable original source, so we have left them out. Before buying, ask a vendor what the tool can do on your own examples, where your data goes and how you can turn it off.
These words are often used interchangeably, but they suit different jobs. Often the simplest option is the right one.
| Tool | How it works | Best for |
|---|---|---|
| Chatbot or assistant | Responds when someone asks | Drafting and answering questions |
| Rule-based automation | Follows fixed steps every time | Predictable tasks with clean data |
| AI agent | Works towards a goal over several steps | Varied inputs where judgement is limited and checkable |
If a task follows the same steps every time, ordinary automation is usually cheaper, easier to test and more predictable. Use an agent when the inputs vary too much for fixed rules.
This takes less than an hour and matches the National AI Centre’s advice to decide who is accountable and to maintain human control.
The subscription is only part of it. Allow for setting up connections, testing, reviewing outputs and maintaining the agent as your tools change. Compare the total with the time saved, using your own before-and-after measurements.
Not for a simple start. Someone still needs to own the process, check the results and understand the data the tool can reach.
Giving an agent broad access and the ability to act before you have seen how it handles real, messy work.
Measure the task before and after: time taken, errors, response speed and how often a person has to correct the output.

One source says 12% of Australian businesses use AI. Another says 72% of small businesses have. Here is what each figure really measures and what it means for you.
Read article
Your first marketing automation should be small, repeated and easy to check. Here is how to choose the right task, set it up and stay within Australian spam rules.
Read article
From 10 December 2026, many Australian organisations must explain how computer programs help make significant decisions. Here is who it affects and what to do.
Read articleGet in touch
A short outline of your goals is a good place to start. Tell us what you want to build, improve or automate.