Automation is leverage applied to a process. If the process is bad, it scales the mistake.
Score four things
Start with volume: how often does the task happen? Then stability: are the rules understood and unlikely to change next week? Add consequence: what happens when the automation is wrong? Finally, recoverability: can a person detect and reverse the mistake?
High-volume, stable and recoverable work is the best starting point. Data entry, notifications, record creation, document preparation and routine routing often fit. Rare judgment calls with serious consequences usually do not.
Fix the process first
Write the current steps and remove unnecessary approvals, duplicate entry and unclear ownership. Automating all the existing steps is not transformation; it is faster bureaucracy.
Keep humans at decision boundaries
Automate gathering, checking and preparing information. Let people approve exceptions, commitments and high-impact changes until the evidence supports moving that boundary.
Measure the outcome
Time saved is useful, but also track error rate, cycle time, exceptions and adoption. An automation nobody trusts creates hidden manual work.
Choose the boring, frequent workflow first. Reliability earns the permission to automate something more ambitious.