What should a small business automate first?

Pick something that happens at least weekly, follows the same rules every time, involves moving information rather than making judgements, and irritates whoever does it. The irritation matters as much as the hours, because the person who resents a task is the person who will help you change it.

Most businesses choose badly on their first attempt, and predictably: they pick the most visible process rather than the most suitable one. The most visible process is usually the most complex, involves the most people, and carries the most exceptions, which means the first automation project becomes a six-month exercise that teaches everyone that automation is difficult. The first one should be almost embarrassingly small, because its real output is not the time saved but the confidence and the pattern.

Four filters do the sorting. Frequency, because a weekly task produces evidence within a fortnight while a quarterly one teaches nothing for months. Consistency, meaning the rules are genuinely the same each time rather than mostly the same with a dozen exceptions. Information-shaped, meaning the work is moving, formatting or routing data rather than deciding something. And an owner who wants it, because enthusiasm survives the awkward first version and mandated projects do not.

Candidates that pass in most Australian small businesses are pleasingly dull. Turning an email enquiry into a record in your CRM or job system. Routing an approval to the right person and chasing it when it sits. Generating the same report every month from data you already hold. Converting a form submission into a task, a folder and a notification. Copying information between two systems that do not talk, which is the single most common source of double entry. Sending appointment or job reminders. Filing documents into the right place based on what they are.

Start with the tools you already own, because most businesses are paying for automation capability they have never opened. Microsoft 365 includes Power Automate, which handles a great many of the above without new licensing. Accounting platforms have rules engines. Job-management systems have triggers and templates. Proving value at zero marginal cost is both faster and far more persuasive internally than a proposal that starts with a purchase.

Measure the way you would measure anything. Time the task honestly for two weeks before changing it, count how often it goes wrong, then measure both again afterwards. Two numbers and one observation, what people stopped doing, is enough. The measurement matters less for the reporting than for the discipline: it forces you to pick a task where the answer is knowable, which rules out the vague ones that waste the most effort.

Tell the people affected before the automation goes live, not after they notice something changed. A task that used to arrive as an email from a colleague and now appears as a system notification is a small change that feels like a large one if it happens without warning, and the instinct is to distrust it. Thirty seconds of explanation converts that into relief, which is the reaction you actually want if you intend to automate anything else.

Then plan for the automation to break, because it will. A form field gets renamed, a permission changes, a vendor updates an interface, and the automation stops without telling anyone. That is genuinely worse than the manual process, because the manual process failed visibly. Build a notification when a run fails, give every automation a named owner, and keep a short list of what exists, since automations built by someone who has since left are a small but real category of business risk.

Write down what each automation does in a sentence, and keep the list somewhere the business can find it. This sounds like bureaucracy for two flows and stops sounding like it at ten, when nobody can remember why a particular email gets filed a particular way. The list also answers the question that arrives eventually during an audit or a handover: what happens to our information automatically, and who decided that.

The honest caveat is that automation amplifies whatever process it encodes. Automating a badly designed approval chain gives you a badly designed approval chain that now runs instantly and is harder to question, because the logic is buried in a tool rather than visible in someone's habit. Spend the first hour asking whether the task should exist at all, since the best automation project is occasionally deleting the step. If you want help finding the first candidate and measuring it properly, call 1800 456 567.