Automate the recurring task that follows the same rules every time, starts from information that is already digital, can be checked quickly by a person, and would save one person about three hours a week. That is usually a chasing, matching or copying job, not the most impressive idea on the whiteboard. To find it, list your recurring admin, score each task on four measures, and pick the highest scorer that clears the three-hour bar.
Why not just copy someone else's list?
Because the right first task depends on your business. A list of ten things to automate can't tell you which one is eating hours in your office this month.
You're not alone in being unsure where to start. In the National AI Centre's SME AI Pulse for December 2025 to February 2026, 19% of SMEs said they don't know how to use AI in their business, and 54% of businesses not adopting it said AI wasn't relevant to them. Most of those businesses still have admin worth automating. They just haven't put a number against each task.
Australia had 2,741,087 actively trading small businesses on 30 June 2026, and businesses with fewer than 20 employees made up 97% of all actively trading businesses. In a team that size there is rarely a spare person to run an automation project. So the first one has to be small, cheap to get wrong and obviously useful.
Step 1: List every recurring admin task
Spend a week writing down admin as it happens. Ask everyone, not just the owner. The person doing the task knows how long it really takes, and it's usually longer than the owner thinks.
For each task, note:
- what triggers it (an email arrives, a date passes, month end)
- who does it and how often
- roughly how long it takes each time
- where the information comes from and where it ends up
Only include tasks that repeat. A job that happens twice a year is not a first automation, however painful it is.
Step 2: Score each task on four measures
Give each task a score from 1 to 3 on each measure.
| Measure | 1 | 2 | 3 |
|---|---|---|---|
| Hours per week (whole team) | Under an hour | One to three hours | Three hours or more |
| How rule-based | Needs judgement each time | Mostly rules, some exceptions | Same steps every time; could be a checklist |
| Cost of today's errors | Annoying, easy to fix | Costs time to unwind | Costs money or client trust |
| Inputs already digital | Paper, phone calls or in someone's head | PDFs and scanned forms | Email, a system, a spreadsheet or a CSV file |
Add up the scores. Then apply one gate before anything else: can a person check the output quickly before it matters? If not, the task drops out of the first round, however high it scores.
A note on error cost. A high score means the task already goes wrong when done by hand, and those mistakes cost something. That's a reason to automate it, as long as a checking step catches what the automation gets wrong. It's a reason not to automate it first if the output goes straight to a client.
Step 3: Apply the three-hour rule
Our three-hour rule says the first build should aim to save one person roughly three hours a week. Much less than that and nobody notices. Much more and the build is usually too big to finish quickly.
Put a dollar figure on it. In our post on what admin costs your business we worked through a person on $95,000 with a 7.6-hour day. An hour a day of their time costs $12,500 a year in wages: $95,000 divided by 7.6.
Three hours a week is three hours out of a 38-hour week, so the same person's three hours cost about $7,500 a year in wages. That's before on-costs, and before the cost of the mistakes the task produces. Run the same sum for each of your top-scoring tasks, using the salary of the person who actually does the work.
Two things to keep in mind:
- Hours spread across several people still count, but the saving is easiest to see when it lands on one person.
- The figure is capacity, not cash. Unless someone's role changes, it buys back time for better work rather than cutting the wage bill.
Step 4: Make sure the result can be checked
The first automation should produce something a person can look over in a minute or two: a list of exceptions, a draft email, a report with totals that tie out. If the only way to know it worked is to redo the task by hand, it isn't saving anything.
This is already common practice. The same National AI Centre survey found that checking AI outputs before they affect customers was the most common safeguard among businesses using AI, in place at about half of them. Build that check in from day one rather than adding it after something goes wrong.
What are good first candidates?
These tasks tend to score well for Australian small businesses because they repeat, follow rules and start from digital information.
- Chasing documents and information. Requests for missing forms, ID, signed authorities or receipts, with reminders until they arrive. A person sees the list of what's still outstanding.
- Appointment and deadline reminders. SMS or email reminders driven by dates you already hold in a calendar or job system.
- Invoice follow-ups. Polite reminders before and after the due date. Payment disputes were the second most common type of dispute the Small Business Ombudsman managed in 2025-26, so steady, early follow-up is worth having.
- Data entry between two systems you already use. A new client in the CRM created in the accounting system, or a finished job turned into an invoice. Retyping is rule-based almost by definition.
- Drafting routine emails for a person to send. Booking confirmations, meeting follow-ups, standard answers to common questions. The system drafts; a person reads and sends.
- Matching statements and reconciliation. Lines on a supplier, insurer or platform statement matched to the bank, with only the mismatches shown to a person. We walk through one version of this in how to automate insurance commission reconciliation.
- Weekly reporting. The report someone rebuilds every Monday from three exports and a spreadsheet.
What should you not automate first?
Some tasks are poor first candidates even when they take real time.
- Judgement-heavy work. Advice, pricing a complicated job, deciding whether a complaint needs escalating. Automation can help prepare this work, but it's the wrong place to start.
- Rare tasks. Year-end jobs and one-off projects. You pay the setup cost once and get the saving once a year.
- Anything where errors reach clients unchecked. If a wrong figure or a badly worded email would go straight to a client, add a review step or pick a different task.
- Tasks with inputs that aren't digital yet. If the information lives on paper or in someone's memory, the first job is getting it into a system.
- Tasks that shouldn't exist. Sometimes the fix is to stop doing the task, or change a form so the information arrives right the first time.
Should you configure or build?
Try configuration first. Plenty of first automations need no custom software at all.
Start with the tools you already pay for. Xero, for example, sends invoice reminders automatically before and after the due date. Booking, practice management and CRM systems often have reminder and workflow settings too. Check what yours can already do before paying for anything new.
Next, connect your tools with Zapier or Make. They work well when:
- the steps are simple and the tools already connect to them
- nothing needs reading, judging or drafting
- a few minutes of setup covers it
That's often enough. If a built-in setting or a Zapier flow saves the three hours, stop there. You don't need a consultant.
A custom build starts to make sense when:
- the inputs are documents someone has to read, such as PDFs, statements or emails with attachments
- the work involves matching data, not just moving it
- several tools need to agree with each other and one of them doesn't connect easily
- the person checking the work needs a proper review screen, not a spreadsheet of results
That's the kind of work we do in AI automation: one task, a fixed price and date agreed in writing before any code, and a person signing off on anything a client sees. We price from what the problem costs the business each year. If the numbers don't stack up, we suggest something smaller, or not building at all.
Frequently asked questions
What is the easiest admin task to automate in a small business?
Usually reminders: appointment reminders, deadline reminders and invoice follow-ups. They run on dates you already hold, follow fixed rules, and many existing tools can do them with a setting.
How do I know if a task is worth automating?
Work out the hours per week and multiply by the cost of the person's time. On a $95,000 salary, an hour a day costs about $12,500 a year in wages. If a task would save one person about three hours a week and the result can be checked quickly, it's a reasonable first candidate.
Should a small business start with AI or simple automation?
Simple automation, if it does the job. Built-in features and tools like Zapier or Make cover most reminders and data copying. AI earns its place when the task involves reading documents, matching messy data or drafting text for a person to review.
What admin should never be fully automated?
Anything a client or regulator sees should be reviewed by a person before it goes out. Judgement calls, such as advice or complaints, should stay with a person, with automation preparing the information.
How long does a first automation take?
Switching on a feature in an existing tool can take an afternoon. Focused custom builds we've quoted have run two to six weeks, with the scope, price and date agreed in writing before work starts.
If you've scored your list and want a second opinion on the top task, book a free 30-minute chat. If a setting or a Zapier flow will do the job, we'll tell you.