A finance team approves supplier invoices. Each invoice takes six minutes of someone’s attention. Invoices arrive steadily through the working day, and the team clears the pile in one run, every day at 4pm.
They automate the step. Six minutes becomes twenty seconds, and it works exactly as promised.
Suppliers notice nothing. Measured end to end, the process is 0.8% faster.
Where did the improvement go?
Answer: Into a queue that was always the real cycle time.
An invoice arriving at 3:59pm is handled a minute later. One arriving at 4:01pm waits until the next day’s run. Across a day of steady arrivals the average wait for the run is twelve hours. Add the six minutes of work and end-to-end is 12h06m.
The automation removed 5 minutes 40 seconds from twelve hours and six minutes. That is the 0.8%.
Now the useful part. Leave the six minutes exactly as they are, un-automated, and process each invoice when it arrives instead of at 4pm. End to end goes from 12h06m to six minutes — a 99% cut, from a change that costs no technology at all.
This is Hammer’s argument in one number. The team optimized the part they could see a way to optimize, and the part that actually held the time was a scheduling convention nobody had questioned since it was set. Before automating any step, find out what fraction of the elapsed time that step occupies. If it is under a few percent, you are polishing something invisible.