1. Does it happen often enough?

Frequency creates both cost and learning. A task that happens dozens of times each week gives you enough examples to understand the pattern and enough future repetitions to justify the work. A quarterly task may still be painful, but the investment is harder to recover and slower to validate.

Count completed instances for a typical month. Then estimate the time spent on the core task, corrections, waiting and chasing. Keep the numbers rough but explicit.

2. Is the route stable enough to describe?

Automation needs a recognisable pattern. The process does not have to be perfectly standard, but the common route and the important exceptions should be nameable. If every case is truly different, first improve the process or build a tool that assists a person rather than trying to run it end to end.

Ask the operator to talk through three normal cases and three awkward ones. Differences that matter will surface quickly.

3. Does the work cross a costly handoff?

Look for copying between systems, repeated re-keying, waiting for approvals and customers being asked for the same context. These joins are useful starting points because they are observable and often have a clear before-and-after state.

A narrow connection between a form, CRM and scheduler can create more value than a broad assistant with no authority or workflow behind it.

4. Can one person own the decision?

A workflow needs an owner who can resolve disagreements, approve the rules and accept the handover. Without that role, the build becomes a debate between departments and the automation inherits every ambiguity in the current process.

Ownership does not mean one person designs alone. Operators, security and customer-facing staff should shape the test cases. It means someone is accountable for the final operating choice.

5. Will you know if it improved?

Choose one or two measures before the build. Time to first useful response, completed intake rate, staff minutes per case and correction rate are more helpful than a general promise of efficiency. Include a quality measure so faster does not quietly become worse.

Score candidate workflows against these five questions. Start with the strongest combination of frequency, stability, handoff cost, ownership and observability. The first release should prove one loop, then earn the right to expand.