Automate the handoff, not the task

A team automates the step people complain about. It genuinely gets faster — ninety seconds becomes four. And the end-to-end time barely moves, because the process was never slow at the steps. It was slow in the gaps between them.

This is the most reliably repeated mistake in automation work, and it is not a technical error. It is a unit error: the work was scoped as a task because a task is what somebody described, and the cost lives at the boundaries where nobody is describing anything.

Measure the waits before you touch the work

Walk one recent instance end to end and write two numbers against every step: how long the work took, and how long it waited before starting. Touch time and queue time.

The ratio is routinely ten or twenty to one in favor of waiting, and once it is on paper the argument changes character. Nobody can seriously propose optimizing four minutes of typing when the same page shows six days of the item sitting in a queue nobody can see.

Use the real instance, not the average. Averages hide the pattern that matters — whether waits are short and consistent, or short most of the time with a long tail that produces every escalation you receive. The tail is usually the business problem, and the tail lives at handoffs.

Four ways a handoff loses time

Notification by hope. The next person finds out there is work because they check, or because somebody mentions it. Nothing fires. Nothing lists it. When they are on leave the item simply stops, and no system anywhere reports that it stopped. This is the most common single failure and the cheapest to fix.

Re-keying. The same information typed into a second system because the two do not talk. It costs the typing, which is small, and it costs correctness, which is not: each re-key is a chance for the two systems to disagree, quietly, in a way somebody will reconcile by hand in three months.

The round trip. Information requested, sent, waited for, received, entered. Four legs, each with its own wait, and the whole loop invisible in any single system — which is why nobody can tell you how long it takes. Ask, and you will get an estimate that is wrong in an interesting direction.

The batch. Work that only moves when somebody runs a list — weekly, or at month end. Average wait is half the batch interval regardless of urgency, so a weekly run means two and a half days of average delay on every item including the ones that matter most.

None of these four is a task. All of them are the connective tissue, and all four are usually cheaper to fix than the step in the middle.

That is the opening. The rest of the lesson — the remaining sections, the exercise and the check-your-understanding questions — is behind the gate below.