The distribution, not the average
The distribution, not the average
"It takes about four minutes" describes a process that does not exist. What exists is two hundred instances, most taking ninety seconds and eleven taking half an hour.
Those eleven are where the pain is, where the automation difficulty is, and where your project will succeed or fail.
Two processes with the same mean
Process A: every instance takes four minutes, without exception.
Process B: 90% take ninety seconds, 10% take twenty-eight minutes.
Both average four minutes. They are completely different problems. A is boring and easy to automate. B has a long tail that probably represents a different task entirely โ the exceptions, the disputes, the ones where somebody has to go and ask.
In B, the 10% is over 70% of the total time. Automating the 90% saves less than a third of the effort and leaves the painful part untouched. That is a genuinely useful thing to know before you build, and the mean hides it completely.
Plot it, do not summarise it
A histogram of your baseline week. Not a mean, not a median โ the actual shape.
Look for a second hump. Most processes that feel painful have one, and it is almost always a different underlying task that has been filed under the same name.
The 90th percentile
The time below which 90% of instances complete. Together with your median from the previous lesson this gives you the shape โ and the ratio between them is the number worth watching.
Median 2 min, p90 24 min
Your baseline week gives a median of 2 minutes and a 90th percentile of 24 minutes. Total volume is 4,000 a month.
The shape of the work
You have just learned something about this process that the people performing it may not have articulated.