What should you automate first?

A task is a good candidate for automation when its boundaries are clear. Frequency alone is not enough: the effect of an incorrect result also matters.

Look for stable inputs

A complete brief, a consistent form or a well-labelled set of records gives a workflow a clearer starting point. If people need to explain the meaning of every input, improve the structure before adding automation.

Consider the cost of a mistake

A draft label can be easy to correct. A published statement or a financial action can have wider consequences. Keep sensitive or consequential decisions under deliberate human review, and avoid giving a process more access than it needs.

Choose an output you can assess

Write down what a useful result looks like. Test ordinary cases and difficult ones, then ask the people using the output whether it helps.

The right first task is often small: organise a request, prepare a summary or pass approved information to the next step. A clear result gives future work a stronger foundation.