Automate first the tasks that are frequent, rule-based, measurable and fed by clean digital data, and where a mistake is cheap to catch. Rare, judgement-heavy or poorly documented work is a bad first candidate, even if it feels painful. A short scoring exercise across your real tasks will show you where to start.
Why a selection method beats a gut feel
The loudest frustration is rarely the best candidate. Teams tend to pick the task that annoys them most, which is often irregular and full of exceptions. A scoring approach pushes you towards work that automation handles reliably.
The NSW Government’s automation guide gives a useful public-sector reference. Its discovery stage suggests building a backlog by asking whether a task is very manual, rule-based with a clear and stable process, low in exceptions, highly repetitive in frequency and volume, prone to human error, measurable and well documented, and fed by standard digital inputs such as spreadsheets, documents, email or PDF (NSW Automation Guide, Stage 1: discover). That list is written for government agencies, so adapt the thresholds to a smaller business rather than copying them.
The four criteria
1. Frequency and volume
How often does the task happen, and how long does each run take? A ten-minute task done forty times a week is a stronger candidate than a three-hour task done twice a year. Measure it for two weeks rather than estimating.
2. Rules
Can you write the steps down so that someone new could follow them without asking questions? If the answer depends on “it depends” or the owner’s instinct, the process is not ready. Document it first. Sometimes the documentation alone fixes half the problem.
3. Error cost
What happens if the automation gets it wrong? Sending a wrong invoice amount is worse than tagging a lead in the wrong list. Low error cost makes a good pilot. High error cost does not rule a task out, but it means a person must review the output. See Human-in-the-Loop Workflows.
4. Data readiness
Automation needs inputs it can read. Check whether the data is already digital, whether it arrives in a consistent format, and whether it lives in one place. The NSW guide also asks you to consider data preparation and suitability, data sensitivity and any added security or compliance needs. If personal information is involved, check your obligations under the Privacy Act and the Australian Privacy Principles before connecting tools.
A simple scoring sheet
List every candidate task and score each criterion from 1 to 5 (5 is best for automation; for error cost, 5 means cheap to catch).
| Task | Frequency | Rules | Error cost | Data ready | Total |
|---|---|---|---|---|---|
| Example: tagging new enquiries by service type | 5 | 4 | 5 | 4 | 18 |
| Example: preparing a bespoke proposal | 2 | 1 | 2 | 2 | 7 |
| Example: weekly sales summary email | 4 | 5 | 4 | 5 | 18 |
The examples above are illustrative only. Replace them with your own tasks and agree the scoring with the people who do the work, not just their managers.
Add context before you commit
The NSW guide’s second step refines the backlog with context: task complexity, upfront and ongoing costs, the need for human oversight, legal compliance, impact on employees, integration with existing systems, scalability and testing. Add a few practical questions:
- Who owns the process and will maintain the automation?
- What system does it touch, and does that system have a stable way to connect?
- What is the manual fallback if the automation fails?
- Will staff understand what changed? Involve them early.
Choose the type of automation
The same guide separates task automation (quick wins from repetitive work), process automation (orchestrating a workflow) and augmentation (giving people intelligent assistance). Starting with task automation builds confidence and gives you real numbers. Rules-based steps are usually better done by plain workflow logic than by AI; reserve AI for the parts that involve reading unstructured text. The wider picture is in How AI Automation Can Improve Business Operations.
Define success before building
For each shortlisted candidate, write one measurable outcome, such as time saved per week or errors each month, and a review date. If the result is unclear, return to How to Assess the Return on AI Projects Without Hype before spending more.
Limits of this method
Scores are judgement, not measurement. Hidden exceptions often appear only once you build. Some valuable work scores low and should still be improved, just not automated first. Treat the list as a way to choose a pilot, and expect to revise it.
Where Radiance Impact fits
Radiance Impact’s Technology team works on workflow automation and ready-made automation kits. You can explore the kit logic in the automation demos, download the free automation checklist, or contact us to talk through your shortlist. The main service overview is at /technology.
Sources
- NSW Automation Guide, Stage 1: discover, Digital NSW
- Automation in context, Digital NSW
