Repetitive work is a good automation candidate when the inputs are known, the rules are reasonably stable and people can explain what should happen when something goes wrong. Examples include moving approved information between systems, preparing routine reports, creating documents from checked data or notifying the next person in a process.
Do not begin with the software. Follow the task from start to finish and record who does each step, what they need, where they retype information and which exceptions require judgement. This often reveals a step that can simply be removed or handled by a feature in an existing product.
The first automation should cover the smallest stable section that produces a useful result. It needs clear checks, useful error messages and a manual fallback. High-consequence decisions—such as releasing money, changing important customer data or making a legal judgement—should not silently pass to an unattended rule.
Measure the current process before promising a saving. Frequency alone is not enough: a rare task may still be worth improving if mistakes are costly, while a frequent two-minute task may not justify a fragile integration.
A useful discovery outcome is a simple process map, a list of worthwhile changes, the product or integration options, and a recommendation for what to do first. It should still help even when the answer is not to build anything.