Make a shortlist from an ordinary week
Write down repeated tasks and estimate how often they happen, how long they take, where they wait, and what correction requires. Include the time someone would spend reviewing the automated result.
Compare frequency, clarity, and consequence
A good first candidate has a clear trigger, consistent inputs, an agreed rule, and a visible result. It also has someone responsible for failures and a manual fallback.
Fictional comparison:
- Internal reminder for an overdue quote: a clear due date and owner make it a reasonable candidate.
- Preparing a document from approved fields: useful if the result is checked before release.
- Sending a final price for an unusual job: poor first choice when scope and pricing need judgment.
Do not start with the most dramatic task just because it makes the best demo.
Try the process manually first
Write the steps, identify missing information, and decide when a person must take over. Check whether your existing tool already supports the rule. If ownership is unclear, fix that before connecting anything.
Set a modest test
Use sample records to check a normal run, a duplicate, and an exception. Then compare a bounded trial against the current process: time spent, corrections, missed items, and review effort. Include maintenance and operating costs.
An estimated time saving is not a guaranteed cash saving. Choose the task where a small, inspectable change can teach you something useful without putting an important decision on autopilot.