Start With a Run Chart: Reading Your Process Before You Control It
When teams first learn statistical process control, they often rush straight to control charts with their upper and lower limits. That is a mistake. Before you can decide whether a process is stable, you need to look at its behaviour over time with fresh eyes, and the run chart is the tool that lets you do exactly that. It asks nothing of you except honesty about your data and a willingness to plot it in order.
A run chart is a line graph of a single measure plotted against time or sequence, with the median drawn across it. That is the whole construction. There are no calculated control limits, no assumptions about distribution, and no statistical software required. In early 2022, with material prices climbing and supplier lead times swinging week to week, this kind of plain visibility has been worth more than any dashboard. A run chart will show you a trend or a shift long before a monthly average ever admits there is a problem.
How to build one
Pick one measure and one cadence. Choose something that matters and that you can collect consistently, such as cycle time per order, defects per shift, or days of inventory on hand. Decide whether you are plotting by hour, day, or batch, and keep it constant.
Plot the points in time order. The horizontal axis is time or sequence; the vertical axis is your measure. Connect the dots so the eye can follow the movement. Do not sort, average away, or reorder the data.
Draw the median. Calculate the median of all the points and draw it as a horizontal reference line. The median, not the mean, is used because it is not dragged around by a single outlier.
Label the context. Note known events on the chart: a new supplier, a staffing change, a return-to-office week. These annotations turn a line into a story you can actually act on.
Reading the signals
The power of the run chart is that a handful of simple, non-random patterns reveal that something other than ordinary variation is at work. You do not need control limits to spot them; you need to count points relative to the median.
A shift: six or more consecutive points all above or all below the median. Points landing exactly on the median are skipped, not counted as breaking the run.
A trend: five or more consecutive points all moving up or all moving down. Repeated identical values are counted once.
Too few or too many runs: count how many times the line crosses the median; a count far outside the expected range hints at non-random behaviour.
An astronomical point: a value so far from the rest that anyone looking would single it out as obviously unusual.
If none of these signals appear, the variation you are seeing is most likely just the normal noise of the process. That is genuinely useful information. It tells you not to react to every up and down, and not to credit or blame a change you happened to make last week. Over-reacting to ordinary variation, what Deming called tampering, makes processes worse, not better.
Why this comes first
A run chart needs no minimum sample size or assumption of normality, so you can start plotting on day one. It is honest about time, which a histogram or a simple average can hide. And it builds the habit of looking at sequence before passing judgement. Once a run chart shows your process is behaving predictably, with no shifts or trends, you have earned the right to move up to a control chart and set statistically grounded limits. Start there too early and your limits will simply be wrong, because you will have baked an unstable period into them.
Treat the run chart as the warm-up that every improvement effort deserves. It costs almost nothing, it travels well on a whiteboard or a single spreadsheet, and it keeps a team grounded in what the data actually says rather than what anyone hopes it says.
If you want help turning raw operating data into clear signals leaders can act on, XNM's strategic advisory can set up the measures and reviews that keep your improvements honest.