Plot It Before You Police It: A Case for Run Charts First
A logistics operation we advised in early 2021 was wrestling with a familiar headache: order-processing time was creeping up, and pandemic-era staffing swings made it hard to tell signal from noise. Their improvement lead wanted to build a control chart on day one — calculate the limits, set the rules, and start flagging out-of-control points. We suggested they hold off and start with something simpler: a run chart. The distinction sounds academic, but getting it wrong wastes weeks.
A run chart is the humble ancestor of the control chart. It plots your measurement on the vertical axis against time or sequence on the horizontal axis, with a centre line at the median. That's it — no control limits, no standard-deviation calculations. Its job is to let you see the behaviour of a process over time before you make any claims about it.
What the run chart revealed
When the team plotted twelve weeks of daily processing times against the median, three things jumped out that a premature control chart would have buried. First, there was an obvious upward trend — a long run of points climbing steadily. Second, there was a sawtooth pattern tied to a weekly batch job. Third, the variation was not stable: the spread in recent weeks was visibly wider than at the start.
That last point is the crux. Control charts assume the process is in a state of statistical control — reasonably stable — before you compute limits. If you calculate control limits on data that is trending or whose variation is still shifting, the limits are meaningless. You end up policing a process against boundaries derived from a process that no longer exists.
Reading a run chart honestly
Run charts come with a small set of well-established tests for non-random patterns. Used carefully, they tell you whether something other than ordinary chance is at work:
A shift. Six or more consecutive points all above or all below the median suggests the process level has genuinely moved.
A trend. Five or more points steadily increasing or decreasing points to a sustained drift, not random wandering.
Too few or too many runs. Counting how often the line crosses the median tells you whether the data is clustering or oscillating more than chance would explain.
An astronomical point. An obvious outlier worth investigating on its own, before it distorts any later calculation.
For the logistics team, the run chart did two useful things at once. It killed the urge to over-react to single bad days, and it told them the process was not yet stable enough for a control chart. They first removed the special cause behind the upward trend — a queue that built up whenever one shift handed off to the next — and only then, once the chart looked stable, did they move on to a control chart to monitor the steadier process going forward.
The practical rule
Start every time-ordered investigation with a run chart; it needs no assumptions and almost no math.
Use a run chart to confirm rough stability before computing control limits.
Watch for trends, shifts, and changing spread — these disqualify a control chart, not just decorate one.
Move to a control chart once the process is stable, when your goal shifts from understanding to ongoing monitoring.
The point is not that control charts are wrong. They are powerful when the conditions are met. The point is sequence: look at the picture honestly first, and let the data tell you whether it is ready for the more demanding tool. Plotting before policing saved this team a month of chasing limits that would have shifted under their feet.
If you want to put real rigour behind your operational metrics rather than reacting to the latest bad number, XNM's strategic advisory can help you choose the right tool for the question you are actually asking.