A stage that took forty minutes on Tuesday and fifty on Wednesday may have changed, or may be doing exactly what it always does. Distinguishing the two is the most useful analytical skill in this field.
Why it matters
Reacting to noise makes things worse. Adjusting a process in response to ordinary variation adds variation.
It wastes attention. Investigating every fluctuation exhausts the people who should be investigating real changes.
It produces false conclusions about interventions, since any change is followed by a return toward the average regardless of whether it worked.
It creates unfair comparisons between shifts and teams, where the difference is ordinary spread.
The two kinds
Common cause: the ordinary variation the process always produces. Present in every measurement, not attributable to anything specific, and reduced only by changing the process.
Special cause: something happened. A machine failed, a large order arrived, a new product line started, half the team was absent.
The response differs completely. Common cause requires process change; special cause requires finding the cause.
Telling them apart
Establish the normal range from a long enough baseline — several weeks covering the usual variation, not one good week.
A point inside the range is ordinary, however unwelcome.
A point outside it, or a run of points on one side of the average, or a steady trend, indicates something has changed.
Plot the series rather than comparing two numbers. Almost every misreading in this field comes from comparing this week to last week without looking at the whole series.
What to plot
The measure over time, as a line, with the average and the normal range marked.
By stage, separately.
Weekly or daily, matched to how often you could act.
One chart per measure, not a dashboard of twelve tiles, which nobody reads.
Reducing variation
Frequently more valuable than improving the average.
A stage with a stable mean and a wide spread is unpredictable, which forces buffers everywhere downstream.
Sources of spread: product mix, equipment differences, method differences between people, exception handling, congestion at particular hours.
Standardising method reduces spread and is what training should target.
Removing the exception path from the main flow reduces spread more than speeding up the main flow.
The comparison trap
Shift A averages better than shift B.
Before concluding anything: check the mix each handled, the volume, the staffing, the equipment available, and the spread.
Then check whether the difference exceeds ordinary variation for the same shift compared with itself over time.
Most shift comparisons that look decisive are within the ordinary range, and acting on them damages morale for no gain.
What to say when asked why a number moved
"Within the normal range" is a complete and correct answer, and it should be given confidently.
Investigate points outside the range, and report what was found.
Report the range itself alongside every figure, so that the question stops being asked about ordinary movement.
Publishing the normal range
The single change that stops noise generating questions.
Compute the ordinary range for each reported measure, from a long baseline.
Mark it on every chart.
State the rule: points inside are ordinary, points outside are investigated.
Answer questions about ordinary movement with "within the normal range", confidently, every time.
Investigate the exceptions and report what was found, which is what makes the rule credible rather than a way of avoiding questions.
Reducing spread rather than the average
Frequently the more valuable target and almost never the stated one.
A stage with a stable mean and wide spread is unpredictable, which forces buffers everywhere downstream.
Sources of spread: product mix, method differences between people, equipment differences, exception handling mixed into the main flow, congestion at particular hours.
Separating the exception path from the main flow reduces spread more than speeding up the main flow.
Standardising method is what training should target, and its effect appears as narrower spread rather than a better average.
Report the spread alongside every average, or this work is invisible.