Volume
Tasks that happen dozens or hundreds of times create more potential value than a process that occurs twice a year. Start by measuring frequency and time spent.
Clarity
The best early candidates have recognisable inputs, an understandable rule or decision and a clear output. If two experienced people disagree on the correct result, full automation is probably premature.
Risk
A wrong internal label can be easy to fix. A wrong payment, legal notice or customer promise can be expensive. Higher-risk steps should include stronger validation and human approval.
Integration effort
A five-minute task is not worth automating if the only available system requires months of brittle integration work. Rank opportunity against implementation cost and maintenance burden.