Control Phase
Purpose
Control makes the gain permanent. The question is not "did it improve?" but "how will we know the moment it slips?"
Elements
Control plan. Names the metric, measurement method, sample size and frequency, chart or check, owner, specification, and the reaction plan (what to do, who to call) when a signal appears. A control plan without an owner or reaction plan is incomplete.
Standard work and documentation. Update procedures, training, and drawings so the new method becomes the default.
Mistake-proofing. Design errors out of the process where possible.
Control charts. Plot data over time with limits computed from the process itself - never from specifications. Choose by data type: variables data uses Xbar-R (small subgroups) or individuals (X-MR); attribute data uses p (proportion defective, varying sample size), np (constant sample size), c (defect count per constant unit), or u (defects per unit with varying unit size). A process can be in control and still off target - capability and control are separate questions.
Common versus special cause. Common cause variation is inherent to the design and is improved only by changing the system. Special cause is a signal - investigate. Adjusting a stable process in response to common cause is tampering and increases variation.
Closure. Verify results against the charter goal, document lessons learned, hand off to the process owner, and report results.
Worked example
Xbar-R chart with n = 5 (A2 = 0.577), grand mean 10.0, average range 2.0: UCL = 10.0 + (0.577 × 2.0) = 11.15; LCL = 8.85. Any point outside those limits is a special-cause signal.
Traps
Using specification limits as control limits, over-adjusting to common cause, and choosing a c-chart when the sample size varies (use u).