Key takeaways
Short answer: Cpk measures process capability using short-term, within-subgroup variation, assuming the process is stable. Ppk uses overall, long-term variation — actual performance, stable or not. If Cpk is much higher than Ppk, the process drifts between subgroups: it is not in statistical control. Reporting one without the other hides whether the process is stable. See also process validation vs process verification.
Cpk is the capability index, calculated from short-term, within-subgroup variation. It describes what the process could do if it were perfectly stable — the best-case index, blind to any drift that happens between subgroups.
Ppk is the performance index, calculated from overall, long-term variation. It includes every shift and drift the process actually experienced, so it describes what you really delivered rather than what you could have delivered.
A process reports Cpk of 1.67 — excellent capability — but Ppk of 1.10. Within any single subgroup the process is tight and capable; across the day it drifts, so the long-term performance is much weaker than the short-term capability suggests. The large gap is the tell: the process is not in statistical control. A team reading only the flattering Cpk would think they were world-class; reading both reveals a stability problem to fix before the capability is even relevant.
Cpk roughly equal to Ppk means a stable, in-control process. Cpk much greater than Ppk means significant between-subgroup variation — the process drifts, and the capability number flatters reality. The gap, not either index alone, tells you whether to chase stability or capability.
Use Cpk to know the process potential, Ppk to know what you actually deliver, and the gap to decide your next move — fix stability (close the gap) or reduce variation (raise both). Reporting only one leaves you blind to half the picture.
1. Reporting Cpk alone. A capable-looking process can still be drifting out of control.
2. Setting control limits from spec. Limits should come from process data, not the spec.
3. Too few subgroups. You cannot estimate long-term variation reliably from a handful of samples.
4. Ignoring the gap. The difference between Cpk and Ppk is the most useful signal.
An unstable process (Cpk much greater than Ppk) produces unpredictable scrap, dragging the OEE Quality rate and making output hard to forecast. Closing the gap stabilises Quality as well as the number.
Fabrico tracks process performance and quality so capability and stability are visible together, not just a single flattering index. Book a demo to see process stability in your quality data.
Usually Cpk — it uses short-term variation and assumes stability.
The process is not in statistical control — it drifts between subgroups.
Both — capability and performance, so the gap is visible.
Enough to estimate long-term variation reliably — a handful is not enough.
Chase stability if the gap is large, or reduce variation if both indices are low.
Login
Blog
