Unit 09.04: Deciding where to stop automating

The value curve flattens and then goes negative. Knowing where is part of the design.

Four slices, decreasing return

Routine, minor variations, real exceptions, and novel cases.

The table below lays them out with a decision each.

slice                                value       decision
the 60% that are routine             clear win   automate
the 25% with minor variations        worth it    automate, more review
the 10% that are real exceptions     marginal    route to a person
the 5% that are novel                negative    do not attempt

85% automated, 15% deliberately not

The third slice is where projects overreach. Ten percent of volume, each case needing a more careful reviewer and a more specific rule - at some point the review costs more than doing the task by hand, and that point is measurable with the numbers from the previous unit.

The fourth slice actively creates work: a longer exception path and a more elaborate guardrail, to cover five percent of items that are all different from each other.

The mistake this prevents

The mistake is treating remaining manual work as a defect. Routing the hard cases to a person deliberately, with a stated boundary, is a finished design. An automation that quietly attempts everything fails unpredictably on the cases nobody chose to include.

Takeaway

Automate the routine slice and route the rest to a person on purpose. Write down where you stopped and why - a stated boundary is what makes it maintainable.

Mark complete