Score each repetitive task on four things — how often it happens, how long it takes, how costly an error is, and how easy it is to automate — then start with whatever scores highest. For most Australian professional firms that first job is a high-frequency, low-risk admin task: appointment reminders, document-request lists, file setup, or follow-up emails. Automate one of those well, prove the time saved, then move up the list. That is the whole method, and the rest of this article shows you how to run it.
This is written for partners, principals and practice managers at Australian accounting firms, law firms, real estate agencies and conveyancers who already know automation would help but cannot decide where to begin. It is the natural next step after you have made a list of candidate tasks — if you have not done that yet, start with what admin work you should automate before hiring another staff member.
Why picking the right first task matters more than the tool
Most automation projects do not fail because the technology is bad. They fail because the firm picked a hard, high-stakes task first, hit a wall, and quietly gave up. Choosing well is the difference between a quick win that builds confidence and a six-week slog that sours everyone on the idea.
The goal of your first automation is not to save the most hours possible. It is to finish, work reliably, and give you something you can point to. Momentum is the real prize early on.
The four-factor scoring framework
Pull together every repetitive task you can think of — anything a person does the same way most weeks. Then score each one from 1 to 5 on four factors.
| Factor | Question to ask | Score 1 (low) | Score 5 (high) |
|---|---|---|---|
| Frequency | How often does this happen? | A few times a year | Many times a day |
| Time | How long does one instance take? | A couple of minutes | An hour or more |
| Error risk | What happens if a human gets it wrong? | Trivial, easily fixed | Costly, hard to undo, or a compliance problem |
| Ease | How clear are the rules and inputs? | Lots of judgement, messy inputs | Clear rules, structured inputs |
Add the four scores for a total out of 20. Frequency and time together tell you how much time you will save. Error risk tells you how much pain the task currently causes. Ease tells you how likely you are to actually finish.
One important nuance on error risk: a high score is a reason to automate, but it also means a human must stay in the loop to check the output. High-risk tasks are worth automating — you just never let them run unattended.
How to read your scores
- Start with your highest total that also scores 4 or 5 on ease. High value plus easy to build is your quick win.
- Park anything that scores low on ease, no matter how valuable. A task full of judgement calls or messy inputs is a project, not a first step. Come back to it once you have a win behind you.
- Treat high error-risk tasks as human-reviewed by default. Automate the preparation, keep the sign-off with a person.
- Ignore anything that is rare and quick. If it happens twice a year and takes five minutes, automating it is a hobby, not a return.
Worked example: an accounting firm
Imagine a three-partner suburban accounting firm during the lead-up to BAS season. Here is how a few candidate tasks might score.
| Task | Freq | Time | Risk | Ease | Total |
|---|---|---|---|---|---|
| BAS and GST lodgement reminders to clients | 5 | 2 | 3 | 5 | 15 |
| Building client tax-return checklists | 4 | 3 | 2 | 5 | 14 |
| Bank reconciliation coding in Xero | 5 | 4 | 4 | 3 | 16 |
| Preparing complex trust distribution advice | 2 | 5 | 5 | 1 | 13 |
Reconciliation scores highest overall, but its ease score of 3 and its error risk of 4 mean it is not an unattended job — automate the obvious matches and review the rest, as covered in AI bank reconciliation in Xero. The genuine first win here is automated BAS and GST reminders: high frequency, clear rules, easy to build, low downside if a reminder goes out a day early. The trust advice, despite taking the longest per instance, is rare and full of judgement — leave it to a person.
Worked example: a law firm
Now a small Brisbane law firm. The temptation is to automate drafting first, because that is where the hours go. The scores usually say otherwise.
| Task | Freq | Time | Risk | Ease | Total |
|---|---|---|---|---|---|
| New client intake and file opening | 4 | 3 | 3 | 4 | 14 |
| Conflict checks | 4 | 2 | 5 | 3 | 14 |
| Drafting bespoke contract clauses | 3 | 5 | 5 | 1 | 14 |
| Converting meeting discussions into action lists | 4 | 2 | 2 | 4 | 12 |
Three tasks tie on total, so ease breaks the tie. Bespoke drafting scores a 1 on ease — the inputs are different every time and an error is serious, so it stays human. The sensible first build is automating client onboarding and file opening: structured inputs, repeatable, and a mistake is annoying rather than catastrophic. Meeting-to-action conversion is a strong low-risk option too — see AI meeting notes for professional services.
Worked example: a real estate agency
A real estate agency's highest-frequency tasks are almost all communication, which makes them ideal first candidates.
| Task | Freq | Time | Risk | Ease | Total |
|---|---|---|---|---|---|
| Open-home follow-up emails | 5 | 2 | 2 | 5 | 14 |
| Drafting listing descriptions | 4 | 3 | 3 | 4 | 14 |
| Rental application screening | 4 | 4 | 5 | 2 | 15 |
| Sorting and triaging the shared inbox | 5 | 3 | 3 | 3 | 14 |
Rental screening tops the table on value, but its low ease score and high error risk — it touches anti-discrimination law — make it a task to approach carefully rather than first, as explained in automating rental application screening without breaking the law. The clean first win is automated open-home follow-up emails: the most frequent task on the list, easy to template, and low consequence if the tone needs a tweak.
Where client data changes the calculation
Error risk and ease are not the only filters. If a task involves personal information — client names, financial details, application data — your scoring has to sit inside your privacy obligations.
The OAIC's general guidance is that the Privacy Act applies the moment personal information is entered into an AI system, and that publicly available chatbots should not be given personal information. In practice that means using your firm's already-licensed, enterprise-grade AI first — Microsoft 365 Copilot inside your tenant, or Gemini in Google Workspace — on enterprise or API tiers where your data is not used for training. Feed the task only the fields it needs, redact identifiers where you can, and keep a written note of where the data flows. A task that would force client data through a consumer chatbot should drop down your list until you can do it safely. We cover this properly in is AI safe for client data under the Australian Privacy Act.
When the answer is "don't automate this"
Some tasks should never top your list, regardless of how many hours they eat:
- Tasks that need judgement every time. Advice, negotiation, strategy. AI can prepare the groundwork, but the decision stays with a person.
- Tasks that must be right 100% of the time with no review. If there is no sensible checkpoint, don't let a machine run it unattended.
- Tasks that happen rarely. The build cost will never pay back.
- Tasks where the process itself is broken. Automating a bad process just produces bad outcomes faster. Fix the process first.
Being honest about these is what keeps your automation programme credible. A firm that automates only the right things trusts the results.
Turning the shortlist into a plan
Once you have your ranked list, don't try to build the top five at once. Take the single highest-scoring, easy-to-build task and implement it end to end. Measure the time it actually saves over a month. Use that result to justify the next one.
Before you commit, it is worth understanding the realistic cost and effort — see how much AI automation costs for a small firm in Australia and our implementation guide for professional services. And whatever you build, make sure your team knows how to use it; training staff to use AI is what turns a clever tool into actual saved hours.
If you want a head start on the shortlist itself, our free guide, 10 AI Workflows to Save 10+ Hours a Week, lays out ten concrete, high-scoring workflows across accounting, legal and real estate — exactly the kind of first tasks this framework points you toward. Download it, score the workflows against your own firm, and you will have your first automation chosen before the end of the day.
Common questions
How do I decide which tasks to automate first in my firm?
Score every repetitive task from 1 to 5 on four factors: how often it happens, how long each instance takes, how costly an error would be, and how easy the rules are to automate. Add them for a total out of 20, then start with the highest-scoring task that also rates easy to build. That combination gives you a quick, reliable win.
Should I automate the task that takes the most time first?
Not necessarily. A task that takes hours but happens rarely, or relies on judgement every time, is a poor first choice because the build cost rarely pays back and the project often stalls. Prioritise frequency and ease for your first automation, then tackle the high-value, harder tasks once you have a working win behind you.
Which tasks should never be automated in a professional firm?
Avoid automating work that needs human judgement each time, such as advice or negotiation; anything that must be perfect with no review step; tasks that happen too rarely to justify the build; and any process that is already broken. Automating a bad process simply produces bad results faster, so fix the workflow before you automate it.
Does automating a task affect our privacy obligations?
Yes. The OAIC's general guidance is that the Privacy Act applies as soon as personal information enters an AI system, and consumer chatbots should not be given personal information. Use your firm's enterprise-grade AI, feed it only the fields a task needs, redact identifiers where possible, and keep a written record of where the data flows. This is general guidance, not legal advice.
How many tasks should I automate at once when starting out?
Just one. Build your highest-scoring, easy-to-implement task end to end, measure the time it saves over about a month, then use that result to justify the next build. Trying to automate several tasks simultaneously spreads your attention thin and makes it harder to prove the value that keeps the programme funded.
Want these ideas working in your firm? We build controlled AI workflows for Australian professional services firms — starting with a free automation audit.