BLOG / OPINION

Correlation isn't cause on the line

Production data often shows strong relationships. But the wrong cause leads to the wrong countermeasure, lost capacity and new instability.

R. Castaño

PLANT TURNAROUND LEAD
23 APR 2026 · 2 MIN READ
opinion

In production, the wrong cause can get very expensive. Not only because the real problem stays put, but because the countermeasure often creates new losses: less throughput, more checks, unnecessary maintenance, wrong holds, or new complexity in the standard.

The risk shows up most when teams treat correlations as causes.

The typical pattern

A quality problem occurs more often when machine A is running. So machine A looks like the cause. The team reduces its use, inspects the machine, schedules maintenance, or reshuffles orders.

Later it turns out: machine A ran mainly with a particular material lot, a particular product family, or a particular shift logic. The machine was visible, but it wasn't causal.

The correlation was real. The conclusion was wrong.

Why factories love correlations

Correlations are fast. They visualise well. They give teams the feeling of having a data basis. And in many cases they're a useful starting point.

But production is a web of dependencies. Parameters influence other parameters. Decisions in the material flow change machine loading. Shifts run differently. Product mix shifts process windows. Maintenance changes act with a time lag.

A correlation doesn't show you which path is behind it.

Causal thinking as a lean extension

Lean has known this problem for a long time. Good A3s separate symptom, direct cause, system cause and countermeasure. Causal thinking makes that discipline more data-capable.

The question isn't: which parameter best predicts the outcome?

The question is: which change would actually move the outcome if we tested it in a controlled way?

What Jidokai can do

On this topic, Jidokai shouldn't show up as a magic cause-finder. Better: as a system that prioritises cause hypotheses more sharply, makes process relationships visible, and protects teams from the obvious spurious correlations.

The decision stays operational: what do we test? What risk does it create? Which measure is reversible? What becomes standard if the hypothesis holds?

The takeaway

Correlation is a hint, not a decision. Take that seriously in the plant and you don't just save analysis time. You prevent countermeasures that leave the real problem untouched.

READ NEXT JidokaiBantō
WRITTEN BY
R. Castaño
PLANT TURNAROUND LEAD

A practitioner perspective from the Lean Competence network, published under a pen name (see our editorial note). Practitioners are available for sprints, fractional and interim engagements.