Exception Triage

Giving the cases that break the flow an owner and a clock

The happy path is the part that already works. It is the small share of cases falling off it that take most of the week, because an exception belongs to whoever noticed it rather than to anyone in particular.

The cost is not the volume. It is that nothing says where a case sits.

The first thing built is a classifier and a queue, not an agent. Every incoming case is scored against the categories the team already uses in conversation, given a reason for the category it landed in, and routed to a named owner with a clock on it. Anything unclear goes straight to a person.

The categories come out of a month of real cases rather than out of a workshop:

• one owner per case, assigned automatically

• a reason attached to every classification

• a clock that starts when the case appears

• an escalation path when the clock runs out

Routing is only useful if the categories are real, which means reading a month of exceptions before writing a single rule and accepting that two of them will turn out to be one, or that the largest is really four in a coat.

Automation comes second, and only where the pattern holds. A category with a stable resolution can be handled end to end; one that still needs judgement gets a prepared case instead, with the history, the likely cause and the next action assembled.

The queue is then the measurement. Volume by category, time to an owner and time to a resolution are visible week by week, so the argument about where to automate next is settled with the queue rather than opinion.

The point is not to make exceptions disappear. It is to make them ordinary work with an owner and a deadline, instead of the thing that quietly consumes everybody’s week.