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

A plain guide to AI for small business Australia: the three jobs AI does well, the one task to start with, and the hype you can safely ignore for now.

Written by Kevin MatthewsHead of Development

Guide5 min read

Start reading
AI for small business, where to start: an inbox where AI drafts customer replies for checking, ranked by risk, with time saved
Start with boring, frequent jobs that are easy to check, like drafting replies you approve.

AI for small business Australia is no longer a future topic. It is a set of tools you can use this month, and most of them are cheap to try. The hard part is not getting access. It is deciding which two or three uses are actually worth your time. This is how we would think about it if you were sitting across the table from us.

What can AI actually do for a small business?

AI does three useful jobs for a small business: it drafts, it sorts and it answers. Drafting means first versions of emails, product descriptions, quotes and social posts. Sorting means reading a pile of enquiries, invoices or reviews and putting each one in the right place. Answering means replying to common questions from customers or staff, at any hour of the day.

Almost everything else you hear about is those three jobs wearing different clothes. A "smart assistant" is usually drafting plus answering. An "automated workflow" is sorting plus a few rules. Holding that picture in your head makes sales pitches much easier to read.

Where should a small business start with AI?

Start with one repeated task that is boring, frequent and low risk. A good first task happens at least weekly, follows a pattern, and has a person who can check the result in under a minute. If a mistake would embarrass you or cost real money, it is not your first task.

Three questions find it quickly:

  1. Does this task repeat in roughly the same shape every time?
  2. Could someone check the result in about a minute?
  3. If the result is wrong, is the worst outcome small and easy to fix?

If you can answer yes, yes and yes, you have found a good candidate. Here is how common tasks usually rank.

Task How risky it is Time to check
Draft a reply to a common customer question Low (you approve before sending) 30 seconds
Write first versions of product or service descriptions Low 1 minute
Summarise a long email thread or meeting notes Low 1 minute
Sort incoming enquiries by type and urgency Medium Spot check daily
Reply to customers automatically with no review High Needs careful setup
Anything involving health, legal or financial advice High Needs expert rules
Table 1Common first AI tasks, ranked by how safe they are to start with.

What should you ignore for now?

Ignore anything that needs you to rebuild your business around it. A good AI tool fits into the way you already work and removes a step. A bad one asks you to change everything first and promises the payoff later.

Here is our short ignore list:

  • "AI agents that run your company". They do not, and the demos hide the supervision behind the scenes.
  • The tool of the week. New products appear constantly, and the basics you learn transfer between them.
  • Any tool that cannot say what it costs per month at your volume. Usage-based pricing can surprise you.
  • Pressure from competitors posting about AI. Posting is not the same as getting results.

How do you set it up without breaking things?

Run the task by hand with AI for two weeks, then automate only the steps that proved themselves. By hand means you copy the input into the AI tool, read what comes back, and fix or send it yourself. It feels slow, but it shows you where the tool is strong, where it makes things up, and what instructions it needs to behave.

After two weeks you will know three things: how often the output was good enough, what mistakes keep repeating, and how many minutes you actually saved. That is the evidence for whether to automate it.

Our rule of thumb is buy before you build. Use an off-the-shelf tool while it fits the way you work. Build something custom when the off-the-shelf version forces you to bend your business around it, or when it cannot connect to the systems you already use, such as HubSpot, Cin7, Stripe or your own database. That is the point where a custom automation starts paying for itself.

Keep a person on the last step

For the first few months, a human approves anything that reaches a customer. The AI drafts, the person sends. Once the draft is right nearly every time, you can loosen that step for the low-risk cases and keep it for the rest.

What does a sensible first month look like?

A sensible first month is one task, one person checking and four weeks of learning. Do not try to fix five things at once. Teams that try end up with five half-working tools and no clear idea which one helped.

1task to start with, not five
2weeks running it by hand before automating
4weeks to a first working use you trust
1person responsible for checking the results

Week one is picking the task and writing down exactly what "good" looks like. Weeks two and three are running it by hand and noting the mistakes. Week four is deciding: keep as is, automate the proven steps, or drop it. Dropping it is a legitimate result. A cheap experiment that teaches you AI is not right for a task has done its job.

If you decide to build something custom, this is where we come in. You speak directly with a lead engineer from the first call to launch, we write the plan down and agree it before any work starts, and we quote on scope and engineering hours, so the same work costs the same price. No templates: whatever we build is engineered for how your business actually runs.

Finished this read?

  1. OpenedToday
  2. FinishedNot yet

Written by

Kevin MatthewsHead of Development

Kevin leads engineering at SingletonTec and has spent 10+ years building production-grade websites, apps and systems for businesses.

Ask Kevin a questionMore from the lead engineer

// questions about this read

Still wondering?Start with these.

Short answers from Kevin and the team. Not here? Ask Kevin directly.

  1. Most small businesses get the best results using AI for three things: drafting text such as emails and descriptions, sorting incoming enquiries or documents, and answering common questions. Start with one repeated task that a person can check quickly, and expand only after it works. If you want help connecting AI to the tools you already use, a lead engineer can map the options with you.

    See custom software and automationTalk to an engineer about thisHow much does AI automation cost in Australia?

  2. Often yes, and small teams can benefit most because a few saved hours a week is a large share of their capacity. The test is simple: does the task repeat, and can someone check the result quickly? If so, even a free or low-cost tool may be worth trying.

    Book a 10-min call

  3. Not to start. Many useful tasks need only a paid account with a general AI tool and a clear set of instructions. You need a developer when you want AI to connect to your other systems, run on its own without someone copying and pasting, or work with your private data safely. That is when a custom setup starts to make sense.

    See custom software and automationAutomate the boring work: a plain guide to business automation (n8n and friends)

  4. Choose the tool that fits the task rather than the one with the loudest marketing. For writing and summarising, a general assistant such as ChatGPT or Claude is a reasonable place to begin. For repeat work across your systems, an automation tool is usually a better fit than a chat window. Whichever you pick, check what it does with your data before you use it for anything customer related.

    AI chatbots for small business: what they do well, what they get wrongTalk to an engineer about this

  5. Not by default. Free and consumer versions of AI tools may store or use what you type, and Australian privacy obligations still apply to personal information you handle. Keep names, health details and financial information out of tools you have not checked, and use business-grade accounts with sensible settings for anything sensitive.

    What not to paste into ChatGPT: AI privacy for Australian businessesTalk to an engineer about this