Cpk, Ppk, and why a capable process can still ship bad parts

Process capability indices compare the room your tolerance gives you against the variation your process actually has. A wide tolerance and a tight process produce a high number. A tight tolerance and a wandering process produce a low one. Cpk and Ppk both do this, and their formulas are nearly identical. The whole difference sits in how they estimate variation, and that difference carries a useful message.

Within versus overall

Cpk uses within-subgroup variation, the spread across a handful of parts made close together under the same conditions. It describes what the process is capable of at its best, when only short-term sources of variation are present. Ppk uses the overall variation across the entire study, which also captures the drift that creeps in over a shift, a day, or a change of material. It describes how the process actually performed.

What the gap tells you

When a process is stable, Cpk and Ppk land close to equal, because the short-term spread and the long-term spread agree. When Ppk falls well below Cpk, that gap is a signal. The process is capable in the moment but drifts over time, and something is changing between subgroups that is worth finding. This is why a single capability number with no context can mislead. A strong Cpk from a quietly drifting process promises a consistency the floor will not deliver.

Reading the value

A common set of thresholds:

• Below 1.00: not capable. The process is producing out-of-tolerance parts.
• 1.00 to 1.33: marginal, capable on paper with little margin for error.
• 1.33 and above: generally considered capable, and many automotive customers ask for 1.33 or 1.67 on important characteristics.
• 2.00 and above: the process uses only half the tolerance, which leaves real room before anything escapes.

Higher is not only about passing. A process that uses a quarter of its tolerance tends to make better-performing product, not merely conforming product, because the parts cluster near the target instead of near the edges.

Before you trust the number

Two conditions sit underneath any capability study, and both are easy to skip:

• The process must be stable. Capability calculated on an out-of-control process describes nothing you can rely on. Confirm stability with a control chart first.
• The measurement system must be trustworthy. If the gauge adds noise, part of the variation you are calling process is really measurement error. A measurement system analysis tells you how much of the spread belongs to the gauge rather than the part.

Capability is one of the most useful numbers in quality and one of the easiest to misread. Confirm the process is stable, confirm the gauge can be trusted, and then let Cpk and Ppk tell you the rest.