01Start with the arithmetic
Take the number of times a task runs per month, multiply by the minutes it takes, and add the cost of the mistakes it produces. Anything under a couple of hours a month is rarely worth an engineering project, however irritating it feels.
The processes worth automating are usually boring and invisible: reconciliations, status updates, data re-entry between two systems that will never be replaced.
02Stability beats volume
A high-volume process that changes every quarter is a worse candidate than a medium-volume process that has run the same way for three years. Automation encodes rules; unstable rules mean permanent rework.
If the process must change often, automate the plumbing and keep the rules in configuration that a non-engineer can edit.
03Automate the path, keep the exceptions human
Aim to cover the predictable majority and route everything else to a person with context. Chasing the last few percent of edge cases is where automation projects go over budget and lose trust.
An exception queue is a feature. It makes the system's limits visible instead of letting it fail quietly.
04Instrument before you celebrate
Measure the cycle time and error rate before the change, then again after. Without a baseline, the value of the work becomes a matter of opinion at exactly the moment you need budget for the next phase.