The process capability index compares the width of your specification limits to the natural spread of your process. The two most common indices are Cp and Cpk, and both use a simple ratio: how much room your specifications allow divided by how much room your process actually uses. A Cpk of 1.33 or higher is the standard benchmark for a capable process in most industries.
The Cp Formula
Cp measures your process’s potential capability, assuming the process is perfectly centered between the upper specification limit (USL) and lower specification limit (LSL). The formula is:
Cp = (USL – LSL) / (6 × σ)
The numerator is your tolerance range, the total spread your customer or design allows. The denominator is six times the process standard deviation (σ), which represents the natural spread of your process. If Cp equals 1.0, the process spread exactly fills the specification window. If Cp equals 2.0, the specification window is twice as wide as the process spread, meaning you have a comfortable margin.
The catch with Cp is that it ignores where your process is centered. A process could be drifting toward one specification limit and still show a high Cp, because Cp only compares widths. That’s where Cpk comes in.
The Cpk Formula
Cpk accounts for how centered your process is by measuring capability toward each specification limit separately, then taking the worse of the two. It uses two sub-indices:
- Cpl = (Process Mean – LSL) / (3 × σ)
- Cpu = (USL – Process Mean) / (3 × σ)
Cpk = the smaller of Cpl or Cpu
Each sub-index measures how many “three-sigma” units fit between your process mean and the nearest spec limit. By taking the minimum, Cpk always reflects the side where your process is closest to going out of spec. If your process is perfectly centered, Cpk equals Cp. If your process drifts off-center, Cpk drops below Cp. The gap between Cp and Cpk tells you exactly how much capability you’re losing due to poor centering.
One-Sided Specifications
Some processes only have one specification limit. A surface roughness might have only an upper limit, or a material strength might have only a lower limit. In these cases, you simply use the relevant sub-index on its own: Cpu for an upper limit only, or Cpl for a lower limit only. There’s no need to calculate Cp, since it requires both limits.
A Worked Example
Suppose you manufacture a metal rod with a target diameter. Your specification limits are USL = 10.05 mm and LSL = 9.95 mm. After collecting data, you find your process mean is 10.01 mm and the process standard deviation is 0.015 mm.
First, calculate Cp:
Cp = (10.05 – 9.95) / (6 × 0.015) = 0.10 / 0.09 = 1.11
Now calculate Cpk:
Cpl = (10.01 – 9.95) / (3 × 0.015) = 0.06 / 0.045 = 1.33
Cpu = (10.05 – 10.01) / (3 × 0.015) = 0.04 / 0.045 = 0.89
Cpk = min(1.33, 0.89) = 0.89
The Cp of 1.11 suggests the process spread could fit within the spec limits, but the Cpk of 0.89 reveals the process is shifted toward the upper limit. You’d need to either re-center the process or reduce variation to bring Cpk above the 1.33 target.
How to Estimate Standard Deviation
The standard deviation you use in these formulas matters more than most people realize. For Cp and Cpk, you want the “within-subgroup” standard deviation, which captures only the short-term, inherent variation in your process. There are two common methods depending on your control chart type.
If you’re using an X-bar and R chart, estimate sigma as:
σ = R̄ / d₂
R̄ is the average range across your subgroups, and d₂ is a constant that depends on your subgroup size. For a subgroup of 5, d₂ is 2.326. For a subgroup of 3, it’s 1.693. These constants come from standard statistical tables.
If you’re using an X-bar and S chart, the estimate is:
σ = S̄ / c₄
S̄ is the average subgroup standard deviation, and c₄ is another constant based on subgroup size. For subgroups of 5, c₄ is 0.9400.
Both methods deliberately exclude variation from shifts and drifts that happen between subgroups over time. That’s what makes Cpk a “short-term” or “potential” capability measure.
Cpk vs. Ppk
You’ll often see Ppk mentioned alongside Cpk, and the distinction is important. Cpk uses the within-subgroup standard deviation described above, which filters out long-term shifts like tool wear, operator changes, or material batch differences. Ppk uses the overall sample standard deviation calculated from all your data points at once, capturing every source of variation in the dataset.
Because Ppk includes more sources of variation, it’s typically lower than Cpk for the same data. Cpk tells you what your process is capable of under stable, short-term conditions. Ppk tells you what your process actually delivered over the full data collection period. If there’s a large gap between the two, your process has significant long-term variation that isn’t showing up within individual subgroups.
How to Interpret Cpk Values
The widely accepted thresholds in quality management are:
- Cpk < 1.0: The process is not capable. It’s producing defects beyond the specification limits.
- Cpk = 1.0: The process barely fits within specs, with the three-sigma boundary touching one specification limit. About 0.27% of output falls outside that limit.
- Cpk ≥ 1.33: The standard minimum target for most manufacturing processes. This provides enough margin that small process shifts won’t immediately cause out-of-spec product.
- Cpk ≥ 1.67: Often required for safety-critical or significant characteristics.
- Cpk ≥ 2.0: A Six Sigma-level process with extremely tight control relative to specifications.
Cpk also translates to sigma levels. A Cpk of 1.0 corresponds to a 3-sigma process (about 66,800 defects per million). A Cpk of 1.33 corresponds to a 4-sigma process (roughly 6,210 defects per million). A Cpk of 2.0 corresponds to a 6-sigma process, yielding only 3.4 defects per million. These conversions assume a 1.5-sigma shift, a standard industry adjustment that accounts for the fact that real processes drift over time.
Requirements Before You Calculate
Process capability indices assume your data meets certain conditions. Ignoring these will produce numbers that look precise but mean nothing.
Your process needs to be stable first. Run a control chart before calculating Cpk. If your process shows out-of-control signals (points beyond control limits, trends, or patterns), the standard deviation estimate won’t be meaningful and neither will your capability index. Fix the instability before assessing capability.
The standard formulas also assume your data follows a normal distribution. You can check this with a normal probability plot or a formal test like the Kolmogorov-Smirnov test. If your data is skewed or follows a different distribution (Weibull, lognormal, etc.), the standard Cpk formula will give misleading results. In those cases, you’ll need to fit the appropriate distribution and use modified capability calculations.
Sample size matters too. NIST recommends at least 100 observations for a reliable capability study. At minimum, you need 25 data points before the estimates have any statistical validity, but 50 or more independent values is a more practical lower bound. With fewer observations, your Cpk estimate will have wide confidence intervals, meaning the true process capability could be substantially different from what you calculated.