A capable process shows variation. Data points drift above and below the target, behaving exactly as expected within control limits. Then, a well-meaning supervisor decides to act. They tweak the temperature, adjust the feed rate, or shift the tool offset to chase the nominal value.
The next data point lands further from the target than the first. So they adjust again. With each intervention, the process—which was stable and predictable—begins to wander. The variation increases, control limits are breached, and defects appear. This phenomenon is not a rare operational failure. It happens in manufacturing plants every day.
W. Edwards Deming demonstrated this counterintuitive truth using the Funnel Experiment. He proved that adjusting a stable process in response to individual results makes the process worse. The mathematics are unambiguous: for a stable process with variance σ², adjusting the process based on the last result doubles the output variance to 2σ².
The Mechanics of the Funnel Experiment
Deming's experiment is simple. A funnel is suspended above a target on a table. You drop a marble through it. The marble hits near the center, scattered by the inherent randomness of the system. This scatter is the voice of the process.
Deming defined four rules for operating the funnel. Rule 1 is to leave the funnel alone. The results scatter around the target with natural, stable variation. This is optimal. Without changing the system itself, this yields the best, most predictable results.
Rule 2 is to adjust the funnel based on the last result. A marble lands right, so you move the funnel left. This compensating action nearly doubles the variation. Rule 3 adjusts based on the distance from the target, which causes a systematic drift away from the optimal setting.
Rule 4 is catastrophic. You place the funnel directly over wherever the last marble landed. The process goes on a random walk, drifting without bound. It is chaos dressed up as responsiveness.
Precision Machining and the Tampering Trap
Consider a precision machining operation producing cylindrical shafts for automotive transmissions. The critical outer diameter is specified at 25.000 mm with a tolerance of ±0.015 mm. The process is capable, boasting a Cpk of 1.67. The control chart shows a stable, predictable pattern.
A newly promoted shift supervisor decides to act. An operator measures a shaft at 25.008 mm. Well within specification, but eight microns above nominal. The supervisor adjusts the tool offset by eight microns to bring it back to center.
The next shaft measures 24.993 mm. Thirteen microns below nominal. Alarmed, the supervisor adjusts the offset by fifteen microns in the positive direction. The next part measures 25.011 mm. The supervisor adjusts again. Within a single shift, a rock-solid process has been transformed into a roller coaster, generating scrap.
The supervisor's report blames unexpected instability. In reality, he was operating under Rule 2 of the funnel. Each individual action was rational in isolation. Together, they doubled the variation.
The Cost of Process Tampering
Distinguishing Signal from Noise
Human brains are pattern-recognition engines wired to respond immediately to deviations. This instinct destroys statistical process control. Individual results from a stable process contain two components: the signal (the true process mean) and the noise (random variation). Adjusting based on a single result means reacting to noise as if it were signal.
This is why Walter Shewhart invented the control chart. It is a decision-making framework that dictates exactly when to act and when to leave the process alone. It separates common cause variation from special cause variation.
Common cause variation is the natural, inherent variability of the process. It is stable and predictable. It cannot be reduced by tweaking the system in response to individual results. It can only be reduced by fundamentally changing the system through better equipment, different materials, or improved engineering.
Special cause variation is unexpected. A point outside the control limits or a run of seven points above the center line signals disruption. These statistical events demand investigation and correction.
Management Tampering and Metric Manipulation
Tampering is not limited to the shop floor. The most destructive interventions happen in management reviews. A KPI shows a slight dip. The executive demands an action plan. Resources are diverted, the process is adjusted, and the next cycle shows improvement—solely because regression to the mean pulled the result back to the process average.
Another management failure is the incentive trap. Tying bonuses to defect reduction targets encourages workers to game the numbers. They redefine what counts as a defect, increase inspection tolerance, or stop reporting borderline cases.
I have audited automotive plants where suppliers promised extreme low PPM defect rates. When actual numbers hovered around 80 PPM, quality managers began reclassifying defects as non-conformances to exclude them from the calculation. The reported PPM dropped, but the customer's incoming defect rate tripled. This is Rule 4 applied at the organizational level.
The moment you react to common cause variation as if it were special cause, you become the source of the variation.
Implementing the Discipline of Inaction
Breaking the tampering habit requires structural changes. First, establish control charts at every critical process and make the chart the trigger for action. If a point is inside the limits and shows no abnormal patterns, no one adjusts anything. This must be an enforced written procedure, not a suggestion.
Second, train operators, supervisors, and managers on the difference between common cause and special cause variation. This is ongoing education. People need to understand that reacting to noise makes the noise louder.
Third, create a decision protocol requiring evidence of special cause before any process adjustment. If an engineer wants to change a setting, they must show control chart evidence justifying the change. No chart, no change.
Fourth, measure the measurement system. Before reacting to any data point, verify the data. MSA (Measurement System Analysis) studies exist for this reason. Adjusting a process based on measurement noise is tampering squared.
Statistical Process Control Decision Flow
- 01Measure the OutputCollect data and plot the point on the SPC chart.
- 02Evaluate Against LimitsDetermine if the result falls inside statistical control limits.
- 03Identify Variation TypeConfirm whether the trigger is common cause noise or special cause signal.
- 04Execute ActionLeave stable processes alone, or initiate a documented 8D investigation.
Building a System That Values Restraint
The hardest discipline for a quality professional to learn is knowing when to do nothing. When your boss asks what you are doing about a data point, explaining that the process is stable often sounds like complacency. It feels counter to every instinct of professional competence.
This discipline separates organizations that improve from organizations that merely oscillate. Improvement requires stepping outside the system, understanding its structure, and making systemic changes.
Celebrate the absence of intervention. When an operator sees a result that is off-target but within control limits and chooses not to adjust, recognize that decision publicly. The culture must value restraint as much as it values action.
The marble does not need your help. The stable process is merely breathing. Leave it alone, and focus your energy on the systemic changes required to actually shift the mean.