The best worker, and why they are not

The previous lesson was about reacting to a process that moved. This one is about the more expensive version: reacting to people who differ.

When several people do the same work in the same system, their results differ. That is guaranteed — identical performers produce different numbers, because the work that reaches them differs. The question is whether the difference between two people is bigger than the difference the system produces on its own.

Almost always, it is not. And almost always, somebody ranks them anyway.

What a ranking of identical things looks like

It looks exactly like a ranking of different things. Somebody is top. Somebody is bottom. The gap between them is real, arithmetically. You can put it in a table and it will be correct.

The tell is what happens next period. A ranking of genuinely different performers is fairly stable: the good ones tend to stay near the top. A ranking of identical performers reshuffles — last month's best is regularly this month's worst, and over a year most people will have visited both ends.

That is not a flaw in the measurement. It is what a ranking of identical things must do.

Deming's demonstration

Deming used to run this with a bowl of 4,000 beads, 20% of them red, and six volunteers he called "willing workers". Each drew 50 beads with a paddle. Red beads were defects.

He would praise whoever drew fewest, put whoever drew most on notice, set a target, and run it again. Then again. He was, deliberately, an entirely reasonable manager doing entirely normal things.

Nobody could influence the result. The paddle takes what it takes. Every difference between the workers, and every difference between days, was the bowl.

The point is not that ranking is unfair. It is that ranking a common-cause system produces information about nothing — and then acts on it, which does real damage:

• The person praised learns that something they did was right. It was not, so they cannot repeat it.
• The person on notice learns that effort does not change the outcome, which is true and demoralising.
• Everyone learns that the measurement is arbitrary, and behaves accordingly.

Targets, and the third option

Set a target below what the system delivers and you have created something that cannot be met by effort. Two things can still meet it: luck, and changing what gets recorded.

The second is not a moral failure of the person. It is the predictable output of an impossible target, and it arrives with a slow degradation of the data everyone else depends on. Once the numbers are being managed rather than measured, no improvement work is possible at all — you have lost the instrument.

What to do instead

Plot the results as one process, on one chart. If everyone falls inside the same limits, there is no difference to explain, and the honest report is "these people are performing the same". That is a real, useful finding.

If somebody genuinely is outside the limits, that is worth investigating — and the first question is still what is different about their work, not about them. Different call mix, different shift, different equipment.

Improve the system, because it is the only thing that will move the number. In the bowl, the only way to get fewer red beads is fewer red beads in the bowl. No amount of attention to the workers changes a 20% bowl.

The experiment opens with six workers over six days.

1. Look at the league table first. It is genuinely persuasive — somebody is best, somebody is worst, and the gap is large. Sit with that for a moment before scrolling.
2. Now read the chart underneath. Every draw inside the limits, and the tool says so in words.
3. Press Run it again. The table completely reorders and the chart says exactly the same thing. Do it three or four times and watch how convincing each new table is.
4. Set the target to 3 and check how many of the 36 draws met it. Then set it to 10 — the process average. Rather more than half now meet it, and the exact count moves every time you reseed, which is the lesson twice over. For this bowl the true figure is 58.4%: the tool counts a draw as meeting a target of 10 when it lands ON 10 as well as below, and 10 is the single most likely result, so including it lifts the pass rate from 44.3% to 58.4%. A target set at the average is met by most people, most of the time, because of where the counting boundary falls — not because of anything they did.
5. Raise days to 15 and look at how many of the workers have led at least once. Given long enough, nearly everybody gets a turn at both ends.
6. Now change the bowl to 5% red. Every worker improves immediately, and none of them did anything. That is the only intervention that ever worked.