An approval screen is easy to design and hard to get used. Our first version was correct in every way that did not matter: it modelled the routing rules precisely, showed the full chain of approvers, and nobody opened it twice.
Sitting with an operations team for a week explained why. An approval is rarely a decision. The decision was already made in a corridor or a call. The approval is a record of it. Designing the screen as a decision point put work in front of people who had already done the work.
Model the workflow as it actually runs, not as the policy document describes it.
So we inverted it. The default view is what is waiting on you, in one line each, approvable without opening anything. The full chain is still there, one click away, for the ten percent of cases where somebody genuinely needs to think. Median time on the screen dropped from ninety seconds to under ten.
The broader lesson is one we keep relearning: model the workflow as it actually runs, not as the policy document describes it. The two are never quite the same, and the gap between them is where software goes unused.
