How do I know whether a process needs AI or needs to change?
Laura Thornton's comments on the accountability panel challenged the idea of adding AI to an unchanged process, and that's a good place to start: inspect the work before you choose the automation. Some steps in a process earn their place and some don't.
Map one request from arrival to completion. Note each point where work passes from one person to another, each approval and each piece of information collected. Beside every step, write what it contributes to the final result.
Some steps are there for a reason
A few protect an important control. Others exist because two systems don't share a field, or because a report was once required and nobody revisited it. Confirm the reason with the people who own the work before you remove anything.
Then sketch a simpler route to the same required outcome. Maybe the person submits the information once, or two reviews happen together, or an outdated report stops altogether. Only after that comparison should you decide what AI would do: organize a request, flag missing information or prepare material for review. It may not be needed for every improvement.
Test the revised process against exceptions as well as the easiest case. Make sure the shorter route still handles the customer, employee or reviewer who needs something different. A faster process is useful when it still produces the right result.
What to try this week
Choose one process your team complains about and write what each step contributes next to it. Mark any step nobody can explain, and ask its owner what it's for.