A better sequence for automation

I think many automation projects begin one step too late. They begin with the technology. A better sequence begins with the work. First, observe what actually happens. Do not rely only on the procedure. Follow real examples. Look for exceptions, waiting, checking, rework and informal coordination. Second, establish what outcome matters. What demand is this work trying to meet? What does good performance look like? What evidence would tell us whether things became better? Third, look for obvious friction. Why is information entered twice? Why does somebody chase another department?

Why does the same error repeatedly require correction? Why does work sit in a queue? Some problems can be removed before we automate anything. Fourth, make judgement and authority visible. Which decisions are routine? Which require context? What could AI support? What could reasonably be delegated? Where should a person remain responsible? Only then choose the technology. That might mean conventional automation. It might mean generative AI. It might mean an agent. It might simply mean redesigning a form or removing an unnecessary approval. Finally, measure what happens to the whole flow of work, not simply the automated activity.

Did lead time fall? Did quality improve? Was capacity released? Did demand stop returning? Did risk change? This sounds slower than beginning with a demonstration and asking where we can deploy it. In practice, a small amount of disciplined investigation at the beginning can prevent months of efficiently implementing the wrong thing. The purpose of understanding work is not to delay action. It is to make action more purposeful.