Intelligent Workflows

The first rule of intelligent workflow design is this: automating a bad process only produces a bad result faster. So we start with process mapping and strip out steps that affect no decision — an approval nobody reads, a field nobody uses, the same data entered into two systems.

In measurement we separate handling time from elapsed time. Processing a request may take forty minutes while completing it takes three days; the difference is waiting, and that is usually where the real cost sits. The goal of design is not only speed but removing wait points: automatic routing, missing-information detection, parallel approvals.

Exceptions are designed up front. Once the exception rate passes twenty percent the user abandons the flow and returns to email, and the automation becomes effectively unused. Every deployed workflow gets volume, elapsed time and error rate indicators and a named owner; ownerless flows degrade over time.