What should you automate first?

Look for friction you can describe.

“We need to use AI” is difficult to scope. “Every Monday, someone copies figures from three systems into the same client report” gives you a starting point. Write down what triggers the work, which systems it touches, and what a finished result looks like.

Use this five-question checklist.

1. Does the work repeat? Look for a task that happens often enough to justify maintaining an automation.
2. Are the inputs available? Identify the actual inbox, form, document, or business record. Check who can access it.
3. Can you describe a correct result? If two people disagree about what should happen, settle the process before automating it.
4. Can someone handle exceptions? Name the person who takes over when data is missing or the request is unusual.
5. Can you measure the starting point? Record time spent, corrections, waiting time, and volume before estimating value.

Keep the first version small.

A useful first scope might collect a request, check required fields, prepare a draft, and send it to a person for approval. Keep irreversible actions behind a clear checkpoint until you have evidence that the process works.

1. 01Request received
2. 02Required fields checked
3. 03Draft prepared
4. 04Person reviews

Work through one example.

Illustrative example, not a client result: a shared inbox receives service requests. A proposed workflow extracts the contact details, flags missing information, and prepares a ticket for review. The first version stops there. Dispatch decisions remain with the team.

Write a one-page workflow brief.

• The trigger and expected volume.
• The systems and access required.
• The definition of a correct outcome.
• The approval and exception owner.
• The baseline and how you will check improvement.

If those answers are still unclear, discovery is the next useful step. A small, well-understood process is easier to evaluate than an ambitious list of disconnected features.