Direct answer
Automate a frequent, rules-based workflow with stable inputs, a clear output, measurable effort and low consequence when an exception is caught. Avoid starting with ambiguous work, sensitive decisions or broken processes nobody owns.
A workflow earns automation in order
High value with low readiness means redesign first, not a faster mess.
Use the first-workflow filter
A first automation should create enough value to matter and be simple enough to learn from. Look for work that happens often, consumes visible time, follows a recognisable path and produces an output the team can verify.
| Question | Strong first candidate | Weak first candidate |
|---|---|---|
| How often? | Daily or weekly | Rare or unpredictable |
| How clear? | Stable inputs, rules and output | Depends on unwritten judgement |
| How risky? | Errors are visible and reversible | Errors create legal, financial or customer harm |
| Who owns it? | Named operator and decision maker | No agreement on the current process |
| Can it be measured? | Time, delay, errors or volume are available | No baseline and no observable result |
Score value and readiness separately
High value does not mean ready. Score each candidate from one to five for frequency, time per item, delay created, error cost, rule clarity, data quality, reversibility and owner readiness. A workflow with high value but low readiness needs redesign before automation.
Use the Advery Workflow Priority Worksheet to estimate recoverable capacity and expose risk. The score is a decision aid, not a business case.
Design the minimum useful workflow
- Write the trigger.
State exactly what event starts the workflow and where that event is recorded.
- Define the required input.
List the fields or documents needed and what happens when they are missing.
- Separate rules from judgement.
Automate deterministic steps first and leave uncertain decisions visible to a person.
- Design the exception queue.
Every failed or uncertain item needs an owner, reason and next action.
- Measure the before and after.
Compare elapsed time, staff effort, rework, missed items and service impact.
Decide whether AI is actually needed
Many useful automations do not need AI. Rules, forms, notifications, integrations and data validation are often cheaper and easier to control. AI becomes useful when the workflow must interpret variable language, summarise material, classify items or draft a response for review.
The Australian Government's Guidance for AI Adoption recommends aligning AI with business goals and managing risk. ASD and ACSC guidance for small businesses also stresses the security risks of cloud-based AI and the need to protect customer data and business systems.
Put controls around the change
- Use the minimum system access and data needed.
- Keep a human approval point for sensitive or consequential actions.
- Log what the workflow did, what failed and who changed the rules.
- Test normal cases, missing data, duplicate events and service outages.
- Define how the team works when the automation is unavailable.
A good first automation creates trust. The team can see what it does, understand its limits and recover when it fails.
Sources and guidance
- Australian Government: Guidance for AI adoption
- ASD ACSC: Artificial intelligence for small business
- OAIC: Privacy and commercially available AI products