Cpk / Process Capability
Compute Cpk / Ppk from your measurement sample and read capability at a glance.
Serves these standards
About this tool
Process capability indices answer one question: can this process hold its specification limits? Cp compares the spread of the process to the tolerance width, while Cpk also accounts for where the process mean sits relative to the target — a process can show an excellent Cp and still fail because it runs off-center.
Run capability studies after the process is statistically stable (see our SPC control chart tool): capability assumes the process is in control, and indices computed on an unstable process are meaningless. As a rule of thumb, Cpk ≥ 1.33 is the common acceptance benchmark, and Cpk ≥ 1.67 is often required for safety-critical or automotive characteristics.
Frequently asked questions
What is a good Cpk value?
Cpk ≥ 1.33 is the widely used acceptance benchmark (roughly 63 ppm defective for a centered, normal process). Many automotive and safety-critical customers require Cpk ≥ 1.67 on significant characteristics. Below 1.00 the process cannot reliably hold specification — recentre it, reduce variation, or use 100% inspection only as temporary containment.
Cpk vs Ppk — what is the difference?
Both use the same formula shape but different spread estimates: Cpk uses the within-subgroup (short-term) standard deviation, while Ppk uses the overall (long-term) standard deviation of all individual values. A large gap between Cpk and Ppk signals drift, tool wear or subgroup-to-subgroup instability that the within-subgroup estimate hides.
How many data points does a capability study need?
Plan for at least 25 subgroups (or 100+ individual readings) collected over a representative production window — material lots, operators and setups included. Fewer points make the standard deviation estimate unstable, and a capability number from an uncontrolled snapshot will not survive a customer or auditor review.
Can Cpk be negative?
Yes. A negative Cpk means the process mean lies outside a specification limit, so a large share of output is already nonconforming. Treat it as a containment emergency: recentre the process first, then rebuild capability.