Look for repetition before novelty
Ask your team where time goes each week. Collect examples of repeated document handling, information lookup or movement between tools.
Choose a task with a clear input and output. If the current process is unclear, map it before trying to automate it.
Keep the first scope manageable
Define one bounded workflow and an owner who understands it. Set a baseline for effort and quality so the pilot has something meaningful to compare against.
Consider whether a simple rule-based automation would solve the problem. AI is one option, not a requirement for every workflow.
Build in review and exceptions
Decide which outputs need human review and what happens when information is missing or uncertain. Keep external actions under appropriate control.
Review what data the tool needs, who can access it and whether the chosen service is suitable for that data. Use representative sample information where possible during early testing.
Learn before expanding
Observe the workflow during ordinary work. A pilot that looks good in a demonstration still needs to fit real habits and edge cases.
Document the process, train its owner and make a deliberate decision about the next stage. Expanding gradually makes it easier to understand what is helping.