Watch how the task actually gets done
A process diagram might show five neat steps. The person doing the work may know about several more.
They check an email because a field is often missing. They ask a colleague because two systems use different names for the same thing. They keep a separate spreadsheet because the main tool does not show the status they need.
These details matter. If you overlook them, an automated version may struggle with the same problems—or make them harder to spot.
Sometimes the first improvement is surprisingly simple: remove a redundant step, collect the right information earlier or decide who owns the handover.
Choose something you can explain
A manageable first automation usually has a recognisable starting point, reasonably consistent information and rules the team can describe.
Consider an incoming request that needs to reach the correct department. If the request contains the necessary details and the routing rules are agreed, there may be a sensible opportunity.
If every request requires a different judgement call, more of the work may need to stay with a person.
That does not rule out automation altogether. It may still help collect information, prepare the task or notify someone that a decision is needed.
Plan for the awkward cases
What happens when information is missing? Or when one system is unavailable?
Someone needs to know that the task has stopped and how to continue it. Agree that before launch, rather than waiting for the first failure.
Afterwards, ask the team whether the change has helped. Has it removed work, reduced delays or simply moved the effort somewhere else?
A small automation that people trust is a better starting point than a complicated workflow nobody feels responsible for maintaining.