Begin with friction, not technology
The strongest automation opportunities usually appear where employees repeat the same coordination work: copying information, finding context, checking completeness, chasing approvals, or rebuilding reports.
Interview the people doing the work and observe real examples. Process documentation alone often misses exceptions, informal handoffs, and workarounds that determine whether an automation will succeed.
Score each candidate
Compare opportunities using the same criteria so a loud request does not outrank a valuable one by default.
- Frequency and time consumed
- Delay or error cost
- Data availability and quality
- Rule clarity and exception rate
- Risk, reversibility, and required oversight
- Measurable benefit within a bounded pilot
Turn a candidate into a testable hypothesis
A useful hypothesis names the users, trigger, allowed inputs, expected output, approval owner, baseline, and success threshold. For example: reducing complete-to-assignment time for qualified leads while maintaining the current routing accuracy.
If the outcome cannot be measured or the source evidence cannot be bounded, the process needs more discovery before it needs an AI tool.