Process capability analysis is a statistical method that measures whether a process can consistently produce output within customer specification limits. It compares the natural spread of the process against the allowed specification width and expresses the result as capability indices — Cp, Cpk, Pp, and Ppk. Cp and Cpk describe short-term capability from within-subgroup variation, while Pp and Ppk describe long-term performance from overall variation.
A higher index means fewer defects, so a Cpk of 1.33 is generally acceptable and 1.67 is often required for critical characteristics. Because it predicts future defect rates, a capability study is usually required before a customer approves a process.
Process capability is one of the most important concepts in quality engineering and manufacturing, yet capability indices like Cp, Cpk, Pp, and Ppk can feel confusing or overly statistical at first. In reality, the core idea is very simple.
Every process has variation. The real question is whether that variation is small enough to consistently meet customer requirements.
Capability metrics help us predict how a process is likely to behave in the future. This is why customers often require a capability study before approving a process, especially for critical characteristics.
In this article, we’ll break down:
- What process capability actually means
- The difference between Cp and Cpk
- The difference between Pp and Ppk
- Why customers care about these metrics
- How short-term and long-term variation affect capability studies
Understanding Specification Limits (LSL and USL)
Specification limits define the acceptable range for a product characteristic. Imagine we manufacture a rubber duck with a target diameter of 100.15 mm. The customer defines these requirements:
- Anything below 100.10 mm is unacceptable
- Anything above 100.20 mm is also unacceptable
These boundaries are called:
- LSL — Lower Specification Limit (100.10 mm)
- USL — Upper Specification Limit (100.20 mm)
Now imagine collecting measurements from the process and plotting them on a histogram. Most measurements cluster around the average, while fewer appear near the extremes, creating the familiar bell-shaped curve of the normal distribution. The key question becomes:
Does the process distribution fit comfortably inside the specification limits?
- If yes, the process is considered capable.
- If no, defects will eventually occur — even if they have not appeared yet.
This is the entire foundation of process capability analysis.
How Many Measurements Does a Capability Study Need?
A capability study needs enough data to be statistically meaningful. Unless the customer specifies otherwise:
- 30 measurements is generally the minimum for meaningful statistical analysis.
- 100 measurements or more is strongly recommended for more reliable results.
In practice, capability analysis should ideally be continuous rather than based on a one-time sample. Manufacturing processes evolve constantly, and a study performed at project launch may no longer represent reality several months later.
Continuous monitoring also adds cost, so companies must balance the cost of monitoring the process against the cost of defects and customer complaints later.
Measurement System Analysis Comes First
Before performing any capability study, the measurement system itself must already be validated. If the measurement system is unreliable, the capability study becomes meaningless.
This is why companies perform MSA (Measurement System Analysis) — to verify whether the measurement system can consistently produce accurate and repeatable measurements. Without a trustworthy measurement system, capability metrics cannot be trusted either.
Mean and Standard Deviation: The Building Blocks
Before understanding capability indices, we need two basic statistical concepts.
Mean
The mean is the average of all measurements. It tells us where the process is centered.
Standard Deviation
The standard deviation measures variation. In simple terms:
- A small standard deviation means measurements stay close together.
- A large standard deviation means measurements are widely spread out.
This matters because customers do not only care about averages. A process can have a perfect average value and still produce defective parts if the variation is too large.
Cp: Potential Process Capability
Cp measures how wide the specification range is compared to the natural spread of the process. It assumes that about 99.73% of output from a stable process falls within plus and minus three standard deviations of the mean — the ±6σ spread that defines the standard capability indices.
Cp = (USL − LSL) / 6σw
Where:
- USL − LSL represents the specification width
- 6σw represents the process spread (six times the within-subgroup standard deviation)
If Cp < 1
The process variation is wider than the specification range. Defects are unavoidable.
If Cp > 1
The process variation is smaller than the specification range. This is better — but Cp alone does not guarantee a capable process.
The Limitation of Cp
Cp assumes the process is perfectly centered between the specification limits. It does not account for a process average that is drifting toward one side, which means a process could already be producing defects while Cp still appears acceptable.
Cp still answers one important question: “If the process were perfectly centered, how capable could it potentially be?”
Cpk: Actual Process Capability
Cpk measures the actual capability of the process, including how well it is centered.
Cpk = min( (USL − μ) / 3σw , (μ − LSL) / 3σw )
Cpk measures the distance from the process mean to each specification limit, then selects the worse side. As a result:
- In a perfectly centered process: Cp = Cpk
- As the process shifts away from center: Cpk becomes smaller than Cp
Mathematically, Cp is always greater than or equal to Cpk. This makes sense because Cp assumes ideal centering, while Cpk reflects reality.
Difference Between Cp and Cpk
- Cp — potential capability: assumes the process is perfectly centered.
- Cpk — actual capability: includes process centering and real-world behavior.
This is why customers usually care much more about Cpk. You can read more in the full article on the difference between Cp and Cpk.
Typical Customer Requirements on Cpk
Each Cpk value maps to a predictable defect rate, which is why customers set Cpk targets rather than judging processes by eye.
| Cpk | Defect Rate (%) | PPM |
|---|---|---|
| 1.00 | 0.27 | 2,700 |
| 1.33 | 0.0063 | 63 |
| 1.67 | 0.000057 | 0.57 |
| 2.00 | 0.000000197 | 0.002 |
Different industries have different expectations, but common Cpk targets include:
- Cpk ≥ 1.33 — generally acceptable for stable production processes
- Cpk ≥ 1.67 — often required for critical or new processes
- Cpk < 1.00 — usually considered incapable
To see how these thresholds map to defect rates in detail, see the guide on Cpk values and sigma levels.
Capability metrics are only meaningful if the process is stable — which is exactly what a control chart verifies. If a process shows a very high Cpk but still experiences defects, that usually points to issues outside normal process variation, such as measurement problems, special causes, assembly issues, human error, or external factors.
Pp and Ppk: Long-Term Performance Metrics
At first glance, Pp and Ppk look very similar to Cp and Cpk, and the formulas are almost identical. The major difference is how standard deviation is calculated:
- Cp and Cpk use σwithin (within-subgroup standard deviation)
- Pp and Ppk use σoverall (overall standard deviation)
Ppk = min( (USL − μ) / 3σo , (μ − LSL) / 3σo )
For a deeper side-by-side comparison, see process capability vs performance (Cp/Cpk vs Pp/Ppk).
Short-Term vs Long-Term Variation
Cp and Cpk — Short-Term Capability
Cp and Cpk use within-subgroup standard deviation. The data is divided into logical subgroups — for example, 5 consecutive parts, 10 parts measured within one hour, or parts from the same mold cavity.
Within each subgroup, conditions are usually similar: same machine, same operator, same material batch, same environment. This lets Cp and Cpk focus mainly on short-term natural variation, so they are considered short-term capability metrics.
Pp and Ppk — Long-Term Performance
Pp and Ppk do not use subgrouping. They calculate standard deviation from all measurements together — the overall standard deviation. This means long-term effects directly influence the variation:
- machine drift,
- operator differences,
- material changes,
- environmental shifts,
- setup variation.
Because of this, Pp and Ppk are considered long-term performance metrics, and they often represent what the customer actually experiences over time.
Long-term variation is usually larger than short-term variation, so Pp is often lower than Cp, and Ppk is often lower than Cpk. If Cp and Cpk look excellent but Pp and Ppk are much lower, the process changes significantly over time and may not be as stable as it initially appears.
Final Thoughts on Process Capability and Performance
Process capability analysis is not just about formulas. It is about understanding how reliably a process can meet customer expectations over time.
The key takeaways:
- Cp measures potential capability
- Cpk measures actual capability
- Pp measures long-term process performance
- Ppk measures long-term actual performance
Most importantly, a good capability study starts with a reliable measurement system and sufficient data. Once these foundations are in place, capability analysis becomes a powerful tool for predicting quality performance, reducing defects, and improving customer confidence. If you want to run your own study, you can calculate Cp, Cpk, Pp, and Ppk for free with the SigmaDesk Process Capability Calculator.
Frequently Asked Questions About Process Capability
What is the difference between Cp and Cpk?
Cp measures potential capability and assumes the process is perfectly centered between the specification limits, while Cpk measures actual capability by including how far the process mean has shifted toward one limit. Cp is always greater than or equal to Cpk, and the two are equal only when the process is perfectly centered.
What is a good Cpk value?
A Cpk of 1.33 or higher is generally acceptable for stable production processes, which corresponds to roughly 63 defective parts per million. Critical or newly launched processes are often required to reach Cpk 1.67 or higher, while a Cpk below 1.00 is usually considered incapable.
What is the difference between Cpk and Ppk?
Cpk uses within-subgroup standard deviation and reflects short-term capability, while Ppk uses overall standard deviation and reflects long-term performance. Because long-term variation includes effects like machine drift, operator differences, and material changes, Ppk is usually lower than Cpk.
Can you calculate process capability for non-normal data?
Not with the standard Cp/Cpk formulas — they assume a normal distribution, so applying them to skewed data can overstate or understate true capability. For non-normal processes, transform the data with a Box-Cox or Johnson transformation, or fit a distribution that matches the data’s actual shape, then calculate capability from the transformed values or fitted percentiles. Checking skewness and kurtosis first tells you whether a transformation is needed.