Cpk and Ppk are two statistical measures that tell you how well a manufacturing process fits within its specification limits. Both compare the spread of your process output against the acceptable range your customer or design requires, but they differ in one critical way: how they measure variation. Cpk captures short-term, “best case” capability, while Ppk reflects actual long-term performance including all the shifts and drifts that happen over time.
What Each Metric Measures
Every manufactured part has specification limits: the highest and lowest values a customer will accept. A shaft might need to be 10 mm in diameter, plus or minus 0.05 mm. The upper specification limit (USL) would be 10.05 mm, and the lower specification limit (LSL) would be 9.95 mm. Both Cpk and Ppk quantify how much of that allowable window your process actually uses, and whether it’s centered within it.
Cpk, the process capability index, tells you what your process is capable of under controlled, stable conditions. It looks at variation within small groups of consecutive parts (subgroups) and ignores the shifts that happen between those groups over hours, days, or weeks. Think of it as the process running at its best. You’ll sometimes see it called “short-term,” “potential,” or “within” capability.
Ppk, the process performance index, uses all measured variation in the dataset, including the shifts between subgroups. It reflects what your process actually delivered over the full observation period. It’s commonly labeled “long-term” or “overall” performance. Because it accounts for more sources of variation, Ppk tends to produce a lower (worse-looking) number than Cpk for the same data.
How the Formulas Work
The formulas for Cpk and Ppk are nearly identical. Each one calculates how far the process mean sits from the nearest spec limit, measured in units of standard deviation, then takes the smaller of the two results:
- Cpk = min[(USL – mean) / (3σ), (mean – LSL) / (3σ)]
- Ppk = min[(USL – mean) / (3s), (mean – LSL) / (3s)]
The only mathematical difference is in the denominator. Cpk uses σ (sigma), a standard deviation estimated from within-subgroup variation using control chart methods. Ppk uses s, the overall sample standard deviation calculated from every individual data point. That single substitution is what separates “capability” from “performance.”
Taking the minimum of the two sides ensures the index reflects the worst case. If your process is perfectly centered between the spec limits, both sides are equal. If it’s shifted toward one limit, the smaller number dominates, pulling the index down.
Why the Gap Between Them Matters
The relationship between Cpk and Ppk is one of the most useful things about tracking both. When a process is stable and in statistical control, there’s very little variation happening between subgroups. The within-subgroup standard deviation and the overall standard deviation are nearly the same, so Cpk and Ppk will be close to each other.
When a process is unstable, the overall standard deviation grows larger because it now includes the between-subgroup shifts (tool wear, temperature swings, material batch changes, operator differences). Cpk stays the same because it only sees the variation within each subgroup, but Ppk drops. A large gap between the two is a red flag: it means your process has significant sources of variation that aren’t visible within short production runs. In one documented example, a process with a Cpk of 1.71 had a Ppk of only 1.48 because of between-subgroup variability. After reducing those shifts and bringing the process into control, the Ppk climbed closer to the Cpk.
In short: if Cpk and Ppk are similar, your process is stable. If Cpk is noticeably higher, your process has hidden variation that needs investigation.
What the Numbers Mean
Industry has settled on widely used benchmarks for interpreting both indices:
- Below 0.67: Not acceptable. The process is producing a large share of out-of-spec parts, often requiring 100% inspection.
- 0.67 to 1.00: Not capable. The process needs improvement.
- 1.00 to 1.33: Insufficient. Parts are mostly in spec, but there’s not enough margin. More frequent inspection is typically needed.
- 1.33 to 1.67: Acceptable. This is the target range for most manufacturing operations.
- Above 1.67: Excellent. The process has so much margin that you may be able to reduce inspection frequency.
A Cpk of 1.67 means the process mean sits five standard deviations from the nearest specification limit, which translates to roughly one defect per million parts. In automotive manufacturing, the AIAG (Automotive Industry Action Group) requires Cpk and Ppk reporting as part of the production part approval process, typically based on a run of at least 300 parts produced under full production conditions.
When to Use Each One
Cpk is most useful during process qualification and when you’re working with control charts. It answers the question: “If this process stays in control, what is it capable of?” That makes it valuable when you’re setting up a new production line, validating a new tool, or proving that a process can theoretically meet spec. It represents potential.
Ppk is the better choice for evaluating what actually happened over a production period. It captures the real-world variation your customers experience, including the effects of material lot changes, shift-to-shift differences, and gradual tool degradation. When a customer or auditor asks “how did your process perform last quarter,” Ppk is the honest answer.
Many quality professionals report both together. Cpk shows what the process could do; Ppk shows what it did do. The combination gives a complete picture. If your Ppk is lower than you’d like but your Cpk is strong, the fix isn’t tightening the process itself. It’s stabilizing it by addressing the sources of shift and drift between subgroups. If both are low, the process has too much inherent variation and needs fundamental improvement.
A Practical Example
Suppose you’re filling bottles to a target of 500 mL, with a lower spec of 495 mL and an upper spec of 505 mL. You collect samples every 30 minutes across a full production day. Within each half-hour sample, the fill volumes are tightly clustered, giving you a small within-subgroup standard deviation. Your Cpk comes out to 1.50, which looks great.
But when you calculate Ppk using the overall standard deviation across the entire day, it drops to 1.10. The difference tells you something is shifting between sampling periods. Maybe the fill nozzle drifts as the supply tank level changes, or temperature fluctuations are affecting viscosity. The process is capable at any given moment but not performing consistently over time. Fixing whatever causes those between-sample shifts would bring Ppk closer to the Cpk of 1.50, giving you a truly capable and stable operation.