A key figure system grows back. Every new problem suggests a new figure, every new manager brings two, and nobody proposes a deletion in a normal meeting. Twice a year the system itself goes on the agenda: not the numbers, but the figures, the targets and the chain.
Five questions per figure
| Question | If the answer is no |
|---|---|
| Did it trigger a decision in six months? | Retire it, or rewrite its purpose and give it two more reviews |
| Is the definition sheet current, with the right owner? | Update it now, in the meeting |
| Is the data source still reliable and the same? | Put it on the reliability check list for this half year |
| Is the target still right after mix changes? | Set a new target with a date, or write why the old one stands |
| Does it still belong at this level? | Move it up or down, and give the level below what feeds it |
Four or five yes means keep. Three means fix. Two or fewer means retire, unless the purpose is rewritten convincingly.
Ninety minutes, four blocks
Fifteen minutes for the four numbers that describe the system rather than the plant: how many figures, how many with an owner, how many triggered an action last quarter, how many are collected by hand. Forty minutes for the five questions per figure. Twenty for the chain. Fifteen for the decisions: what is retired, what enters in its place, which targets are reset.
The entry rule
A system without a fixed size grows back whatever the review decides. The top level has a number of slots, agreed by the plant manager; in the example plant it is 24. A new figure enters only when a named figure leaves. Each figure has one person who explains it and proposes its retirement when it stops steering.
Without a ceiling, the comparison that matters never happens: is this new figure better than the weakest one we already have?
What the reviews did in the example plant
| Review | Retired | Added | Size after |
|---|---|---|---|
| First cut, month 2 | 118 | 0 | 28 |
| Half year, month 6 | 6 | 2 | 24 |
| Year end, month 12 | 1 | 1 | 24 |
The second review proves the habit. A plant that cuts hard once and never prunes again is back at 60 figures within three years.
Checking the chain
The chain is checked link by link, with the actual values next to the expected ones. Skills coverage to unplanned downtime: expected 85 and 6.0 percent, actual 84 and 5.6. Held. Downtime and OEE to lead time: expected 70 percent and 8.0 days, actual 70.1 and 7.8. Held. Lead time to on-time delivery: expected 95.0 percent, actual 95.4. Held. Output to conversion cost per unit: expected and actual EUR 44.00. Held.
Output to energy cost per unit: expected EUR 1.95, actual EUR 2.05. Failed, because the extra output came from a line with a higher connected load. That one failed link produced the only genuinely new insight of the year.
The hardest part
Retiring a figure feels like admitting it was never needed. It is the opposite: a system that never loses a figure has stopped learning. Count retirements as a result of the review and report them next to the improvements.
What the system looked like after twelve months
146 figures became 24. Owners went from 69 to 100 percent, figures that triggered an action in a quarter from 22 to 83 percent, manual collection from 61 to 29 percent. The monthly report fell from 38 pages to 4, the review from 75 minutes to 40. On-time delivery reached 95.4 percent and conversion cost per unit EUR 44.00.
None of that came from a better dashboard. It came from asking, for every number, which decision it supports and who makes it.