If you’ve ever opened a capability report in Ellistat looking for the Cpk, you didn’t find it. This isn’t an oversight. It’s a deliberate choice, based on a simple conviction: in mass production, the Cpk almost always paints a more flattering picture of the process than what the customer actually experiences.
Cpk and Ppk are so similar that they are often confused. However, they are based on two very different perspectives on variability. Understanding this difference helps prevent validating a process based on its best performance rather than its typical performance. Here’s what sets these two metrics apart, why we’ve made PPK our benchmark, and how to use it correctly.
Cpk and Ppk: Same Formula, Different Variance
Both indices measure a process's ability to produce within tolerances. They combine two pieces of information: the dispersion of the parts and the position of the mean relative to the limits.
Their formula is the same: you take the distance between the mean and the nearest tolerance limit, and divide it by three times the standard deviation. The only difference lies in the standard deviation used.
CPK: Short-Term Capability
CPK uses a standard deviation short term, estimated based on the variation within the subgroups (using the mean range or mean standard deviation of the subgroups). Since samples from the same subgroup are collected consecutively, this estimate does not account for anything that occurs between two samplings.
It assumes that the process is under statistical control: stable, with no drift and no special causes. The Cpk therefore describes the potential of the process under ideal conditions.
The PPK: Long-Term Performance
The PPK uses the standard deviation global, calculated based on all the values for the observed period. It incorporates all sources of actual variation:
- gradual wear and tear on tools;
- changes in raw material batches;
- differences between teams or operators;
- machine restarts, temperature fluctuations;
- successive corrections and adjustments.
The PPK describes the actual performance of the process, whichever is noted during the final inspection or by the customer.
An example to help clarify the concept
Let's take a measurement of 10 ± 0.05 mm, with a mean of 10.01 mm. The nearest limit is 0.04 mm from the mean. (For illustrative purposes only.)
- Short-term standard deviation: 0.008 mm → Cpk = 0.04 / 0.024 ≈ 1.67
- Overall standard deviation over one month: 0.012 mm → Ppk = 0.04 / 0.036 ≈ 1.11
Same process, same parts, same alignment. On one hand, a process that seems more than capable; on the other, a process that is barely acceptable. The only difference: one looks at a few consecutive parts, while the other looks at the production as it actually took place.
Why Cpk Overestimates Capability in Mass Production
CPK is a useful tool in its field: process development, initial qualification, and controlled environments. It helps isolate the intrinsic variability of the machine and the process.
The problem arises when it becomes the standard indicator in production. A production line never operates for long under perfectly stable conditions. Tools wear out, material batches change, and the machine on Monday morning doesn’t behave the same way as it does on Wednesday afternoon. These phenomena aren’t anomalies—they’re the normal conditions of industrial production.
However, by design, the Cpk ignores these factors. A process that drifts slowly but remains highly repeatable between two consecutive parts will show an excellent Cpk, even though it actually produces parts with much greater variation.
Managing by Cpk therefore involves two risks:
- validate a process that is not capable under real-world conditions;
- to discover the problem too late, during the final inspection or at the customer's site.
The PPK, Ellistat's benchmark indicator
A capability indicator must answer a specific question: Does my process actually produce conforming parts under real-world conditions over a representative period of time?
The PPK provides a direct answer. Calculated using long-term data collected under normal operating conditions, it predicts what the customer will see. A PPK of 1.33 over three months of production is a solid indicator. A Cpk of 1.67 measured over two hours on a Tuesday morning is a fragile promise.
It is also the metric we use to measure our own performance. For a customer using the APC, the Ppk rose from less than 1 to about 1.4: an improvement achieved based on actual production—with all its uncertainties—rather than on a selected time window.
What if your customer requires a Cpk?
This is the most common objection, particularly in programs where validation applications explicitly require a Cpk.
The answer is simple: providing a PPK means providing an indicator more demanding. Since the overall dispersion is, in practice, greater than or equal to the short-term dispersion, a process that meets the required threshold for Ppk also meets it for Cpk. The reverse is not true.
In practical terms, presenting a Ppk to your client is equivalent to telling them: «Here is how my process performs under real-world conditions, including changes in shifts and batches.» This is a point that builds trust—not a weakness in your case. Include the data and the time period covered so the client can assess how representative they are.
Diagnosing Without Using Cpk as a Control Metric
It is still useful to distinguish between short-term and long-term variability, particularly when diagnosing a problem: Is the process inherently too variable, or is it instability over time that is degrading performance?
The right tool for this isn't a Cpk value displayed as part of a routine, but the control chart :
- The range (or standard deviation) map shows the dispersion within the subgroups—that is, the short term;
- The map of averages shows the deviations, jumps, and shifts between the subgroups.
If the intra-subgroup variation is low but the averages fluctuate significantly, the process is capable but unstable: efforts must focus on monitoring and correcting deviations. If the intra-subgroup variation is already high, it is the process itself (machine, tooling, production sequence) that needs to be improved.
How to Collect Valid Data for a PPK
The PPK is only meaningful if the data is representative. One hour of stable production is not enough. A good basis for calculation includes:
- several teams and, if possible, several operators;
- multiple batches of material ;
- several tool changes and machine restarts;
- one significant period in terms of frequency (at least one week in a row, often more).
Two precautions round out this data collection plan. First, verify the capability of the measurement method: high measurement variability artificially inflates the overall standard deviation and negatively affects the Ppk. Second, examine the shape of the distribution: a Ppk calculated under the assumption of normality for data that is not normally distributed can be misleading.
This rigor requires a little more work up front. In return, it provides an accurate picture of what the process actually does, and it is the only solid basis for deciding whether it is under control or needs improvement.
From Assessment to Action: Improving the Ppk
Measuring an accurate Ppk is the first step. Improving it is another. In most machine shops, the gap between potential and actual performance stems from variations such as tool wear, thermal expansion, and manual corrections that are made too late or are excessive.
Two factors directly influence this gap:
- the SPC, which makes deviations visible in real time using control charts and allows for action to be taken before nonconformity occurs;
- the APC, which automatically corrects the machine in a closed-loop system based on measurements, using machine learning to prevent overcorrection. It is precisely the long-term variability measured by the Ppk that is reduced.
Conclusion: Focus on reality, not potential
Cpk and Ppk share the same formula but not the same view of the process. Cpk describes what the process could do under ideal conditions; Ppk describes what it actually does. At Ellistat, we have chosen PPk as our capability indicator because it is the one that predicts the quality delivered to the customer. Combined with control charts for diagnosis and data collected over a representative period, it provides a reliable basis for decision-making, investment, and improvement.
FAQ
What is the difference between Cpk and Ppk?
Both use the same formula. Cpk is based on the short-term standard deviation, estimated within the subgroups; Ppk is based on the overall standard deviation of all data for the period. Cpk measures the process’s potential, while Ppk measures its actual performance.
Why is the PPK generally lower than the Cpk?
Because the overall standard deviation accounts for all sources of variation over time (drift, material batches, teams, settings), whereas the short-term standard deviation considers only the variation between consecutive parts. The more unstable the process, the greater the difference between the two.
What Ppk value should be targeted?
A Ppk of 1.33 is the threshold commonly used for a standard characteristic. Critical characteristics or certain customer specifications require higher thresholds, often 1.67. The exact threshold must be determined in consultation with the customer.
How much data is needed to calculate a reliable Ppk?
There is no universal number. The key is that the data cover the actual sources of variation: multiple teams, multiple material batches, and multiple tool changes, over a period representative of production.
What should I do if my client asks for a Cpk?
Present the Ppk to him, along with the data and the period covered. In practice, a process that meets the Ppk threshold also meets the Cpk threshold, since the overall variation is almost always greater than or equal to the short-term variation.
