Cp and Cpk are often displayed next to each other, but they answer different questions. A useful review starts by separating three ideas: the width of the specification, the spread of the process, and the location of the process mean.
The basic definitions
For a two-sided specification, Cp compares the specification width with the estimated process spread. In the familiar six-sigma form:
Cp = (USL − LSL) / (6σ)
Cp does not include the distance between the process mean and the nearest specification limit. It therefore describes a potential capability: how the process spread would fit if the process were perfectly centered.
Cpk takes the smaller of the upper and lower one-sided capability values:
Cpk = min[(USL − mean)/(3σ), (mean − LSL)/(3σ)]
Because it uses the nearest limit, Cpk falls when the process mean moves away from the centre of the tolerance.
What the gap between Cp and Cpk means
| Pattern | Practical interpretation |
|---|---|
| Cp and Cpk are similar | The process mean is approximately centred between the two limits. |
| Cp is clearly higher than Cpk | The process spread may fit, but the mean is shifted toward one specification limit. |
| Both are low | The process spread is too large for the tolerance, and centering may also be a problem. |
A large Cp–Cpk gap often suggests a centering opportunity, but the correct action depends on the process. Adjusting a mean without understanding tool wear, drift, setup logic, or safety margin can create a new problem.
Capability should follow stability
Capability indices summarise a distribution relative to specifications. If the process is changing over time because of special causes, a single mean and standard deviation may not represent the future process. That is why a control-chart review is normally separated from a capability review.
Before presenting a final Cpk conclusion, check the sequence of measurements, shifts, lots, machines, cavities, operators, material batches, and other known sources. A dataset that combines different process streams may produce a smooth histogram while hiding the real structure.
Which standard deviation is being used?
Different software and procedures may distinguish within-subgroup and overall standard deviation. This leads to paired indices such as Cp/Cpk and Pp/Ppk. The exact method should be visible in the report. Comparing numbers from different sigma estimates without noticing the method can create an apparent disagreement that is only a definition difference.
Sample size and sampling logic matter
A capability estimate from a small or selectively sampled dataset has greater uncertainty. More rows do not automatically solve the problem if the rows cover only one short period or exclude known variation. Sampling should represent the conditions the capability statement is intended to cover.
Common mistakes
- Using customer specification limits as control limits.
- Calculating Cpk before checking process stability.
- Combining machines, cavities, or product families without stratification.
- Removing outliers without a documented technical reason.
- Using rounded or transformed data without understanding the effect.
- Treating a threshold as an automatic process approval.
A practical review sequence
- Confirm the characteristic, unit, LSL, nominal, and USL.
- Check the measurement system and data resolution.
- Plot measurements in time or production order.
- Stratify by known process sources.
- Review stability using an appropriate chart.
- Calculate capability with the approved sigma method.
- State the sampling period, sample size, exclusions, and conclusion limits.
Use the number as evidence, not as the entire decision. A capability index should support a documented engineering conclusion that includes the process state and the data limitations.