The Control phase is where Six Sigma projects most often fail quietly. The improvement was real, the tollgate was passed, the team moved on, and eighteen months later the metric is back where it started. The Control phase exists to prevent that, and the control chart is its central instrument. This article explains what control charts do in the Control phase specifically, how they differ from their use in earlier phases, and what a control-phase chart needs around it to actually hold a gain.
The purpose of the Control phase
Define, Measure, Analyze, and Improve produce a change to the process. Control makes the change permanent. It has three jobs:
Verify that the improved process is stable at its new level.
Monitor it so that any drift back toward the old level is detected quickly.
Transfer ownership from the project team to the people who run the process day to day.
The control chart serves all three, but it does so only when it's set up for the improved process rather than carried over from the baseline.
Control charts earlier in DMAIC
To see what's different about Control, it helps to see how the chart is used before it.
In Measure, a control chart on the baseline data answers the question "is the process stable enough to characterize?" If it isn't, the capability figures are meaningless and the Analyze phase will be chasing a moving target. The chart's job here is diagnostic.
In Analyze, charts are often stratified (by shift, machine, lot, operator) to see whether the special causes cluster in a way that points to a root cause. The chart is an investigative tool.
In Improve, a chart on pilot data shows whether the change shifted the mean or reduced variation, and whether the pilot process is stable. The chart is evidence.
In Control, the chart becomes an operational tool. It's no longer analyzed by the project team; it's watched by the operator, and it triggers a defined response. That shift in purpose changes how it should be built.
Building the control-phase chart
Recompute the limits from the improved process
The most common Control-phase mistake is keeping the baseline limits. If the project reduced variation, the old limits are too wide and the chart will never signal a regression until the process has fully reverted. If the project shifted the mean, the old center line is wrong and every point will look like a signal.
Collect 20–25 subgroups from the improved process, confirm stability, and compute new limits. Those become the monitoring limits. Document them in the control plan with the date and the data they were computed from.
Choose the chart for the operator, not the statistician
A Control-phase chart is read by someone whose main job is running the process. Choose the chart type and subgrouping accordingly:
If measurement is fast, an X̄-R chart with n = 3–5 per hour is standard.
If measurement is slow or destructive, an I-MR chart with one measurement per interval is more realistic.
If the metric is a defect rate or count, a p or u chart.
If the improvement was about a small sustained shift and detection speed matters, consider an EWMA chart alongside the Shewhart chart, but only if the operator will be trained to read it.
The best chart is the one that will actually be maintained. A technically superior chart that requires ten minutes of measurement every fifteen minutes will be abandoned.
Keep the rule set small
Apply the one-point-outside-limits rule, a run of eight, and possibly a trend of six. More rules raise the false alarm rate and lower trust in the chart. The Control-phase chart is for catching regressions, not for research.
The reaction plan
A control chart without a reaction plan is a graph. The reaction plan is a short written document, attached to the chart, that says:
What constitutes a signal on this chart (the rules in use).
What to check first when a signal occurs, in order. This is where the project's root-cause findings pay off: if the Analyze phase found that the dominant cause was fixture wear, the first check is the fixture.
When to stop the process versus continue while investigating.
Who to notify and within what time.
How to record the signal, the investigation, and the outcome.
The reaction plan converts the project team's knowledge into the operator's procedure. Without it, the operator either ignores the signal or adjusts the process blindly, which is tampering.
The control plan
Six Sigma projects typically produce a control plan document that lists, for each critical characteristic: the specification, the measurement method and frequency, the control chart type and limits, the reaction plan, and the owner. The control chart is one row in that table. The control plan is what the process owner signs at the tollgate, and it's the artifact an auditor will ask for.
The control plan should also state when the limits will be reviewed. A common convention: review after 25 subgroups of monitoring data, and again whenever the process is deliberately changed. Limits are not permanent.
Handoff
The Control phase ends when the process owner accepts responsibility. Practically, that means:
Operators have been trained on the chart and the reaction plan and have demonstrated they can respond to a signal.
The chart is being maintained without project team involvement.
At least a few weeks of monitoring data show the improved process holding.
The control plan is signed.
A useful test: if the project lead disappeared tomorrow, would the chart still be plotted next week and would a signal still trigger the right response? If not, the phase isn't finished.
Common Control-phase failures
Charts that are plotted but never read. Data are entered, the chart updates, nobody looks. The fix is a reaction plan with a named owner and a periodic review in which someone checks whether signals were investigated.
Limits inherited from the baseline. Discussed above. The chart can't detect regression toward a state its limits already include.
Charting an output that's too far downstream. If the improvement was to a setup procedure and the chart tracks final inspection defects, a regression will show up days later after hundreds of parts. Chart the characteristic closest to the change.
Charts that signal constantly. Either the improved process isn't actually stable, in which case the project isn't done, or the limits were computed on the wrong variation. Either way, a chart that signals every shift will be ignored within a month.
No mechanism to recompute limits. Processes are changed. When they are, the limits must be redone, and there needs to be a defined trigger for that.
Treating the chart as proof of capability. A stable chart says the process is predictable. It says nothing about whether it meets specification. The control plan should include the capability index computed from the chart's σ estimate, and that figure, not the chart alone, is the evidence that the improvement met its goal.
Tools
The Control phase doesn't need sophisticated software. It needs a chart that operators can update and read on the floor. A browser-based control chart builder that computes limits from the improved-process subgroups and flags rule violations is sufficient for most projects, and it has the advantage of not requiring a statistics package license at every workstation.
For the conceptual foundation that operators need to read the chart correctly, especially the distinction between common cause variation (leave it alone) and special cause variation (follow the reaction plan), the what is a control chart overview is a good training reference.
Summary
In the Control phase, the control chart changes from an analytical tool to an operational one. Recompute the limits from the improved process, choose a chart the operator will maintain, keep the rule set small, attach a written reaction plan, embed it in the control plan, and hand it off to a named owner. The chart's job is to detect regression early enough that the gain is protected; everything around it exists to make sure someone acts when it does.