The process capability index compares the spread of your process output to the width of your specification limits, giving you a single number that describes how well your process fits within tolerance. The most common version, Cp, is calculated by dividing the specification width (USL minus LSL) by six times the process standard deviation. A value of 1.33 or higher is the widely accepted minimum for a capable process.
The Cp Formula
Cp measures your process’s potential capability, assuming the output is perfectly centered between specification limits. The formula is:
Cp = (USL – LSL) / 6σ
USL is your upper specification limit, LSL is your lower specification limit, and σ is the process standard deviation. The denominator (6σ) represents the natural spread of a normal distribution, covering 99.73% of output. If your specification range is exactly six standard deviations wide, Cp equals 1.0, meaning your process barely fits inside the limits with almost no room for error.
For example, if your USL is 10.0 mm, your LSL is 9.0 mm, and your process standard deviation is 0.15 mm, the calculation is: (10.0 – 9.0) / (6 × 0.15) = 1.0 / 0.9 = 1.11. That Cp of 1.11 tells you the process spread is narrower than the spec range, but not by a comfortable margin.
Why Cp Alone Isn’t Enough
Cp only measures spread. It assumes your process mean sits exactly at the midpoint of your specification range. In practice, processes drift. A process can have a great Cp but still produce defects because its center has shifted toward one specification limit. This is where Cpk comes in.
The Cpk Formula
Cpk accounts for both spread and centering by measuring the distance from the process mean to the nearest specification limit. It’s calculated as the smaller of two values:
Cpk = minimum of (Cpu, Cpl)
Where:
- Cpu = (USL – process mean) / 3σ
- Cpl = (process mean – LSL) / 3σ
Cpu tells you how capable the process is relative to the upper limit, and Cpl tells you the same for the lower limit. Taking the minimum gives you the worst-case side, which is the one most likely to produce out-of-spec parts.
Using the same example (USL = 10.0, LSL = 9.0, σ = 0.15), but now the process mean is 9.7 instead of the ideal 9.5: Cpu = (10.0 – 9.7) / (3 × 0.15) = 0.3 / 0.45 = 0.67. Cpl = (9.7 – 9.0) / (3 × 0.15) = 0.7 / 0.45 = 1.56. Cpk = min(0.67, 1.56) = 0.67. Despite a Cp of 1.11, the off-center mean drops the Cpk to 0.67, which is well below acceptable. The process is too close to the upper limit.
When Cpk equals Cp, the process is perfectly centered. When Cpk is lower than Cp, the process has shifted off-center, and the gap between the two numbers tells you how much.
Step-by-Step Calculation
Before plugging numbers into a formula, you need to verify a few things. The capability index assumes your process is stable (in statistical control) and that your data follows a roughly normal distribution. NIST recommends collecting at least 50 independent data points for reliable estimates. If your data isn’t normally distributed, you can apply a transformation like the Box-Cox method to make it approximately normal before calculating.
Once those conditions are met:
- Step 1: Identify your specification limits (USL and LSL) from the product or engineering requirements.
- Step 2: Calculate the process mean from your sample data by averaging all measured values.
- Step 3: Calculate the process standard deviation using the sample standard deviation formula.
- Step 4: Calculate Cp by dividing the spec width by 6σ.
- Step 5: Calculate Cpu and Cpl using the process mean, then take the minimum to get Cpk.
If you’re working with subgroups (small batches measured at regular intervals), you can estimate the standard deviation from the average range of those subgroups divided by a statistical constant (known as d2, which depends on your subgroup size). This within-subgroup method is the standard approach for Cp and Cpk because it captures short-term variation only.
One-Sided Specification Limits
Some characteristics have only one specification limit. Surface roughness might have only a maximum, for instance, or a strength measurement might have only a minimum. When only one limit exists, you can’t calculate Cp because there’s no tolerance width to measure. In that case, you only calculate the relevant one-sided index: Cpu if you have an upper limit, or Cpl if you have a lower limit. That single value serves as your Cpk.
What the Numbers Mean
Capability index values map directly to sigma levels and expected defect rates. Here are the key benchmarks used across most industries:
- Cp or Cpk = 1.0: A 3-sigma process. The specification limits line up exactly with the natural process spread. Roughly 2,700 defects per million opportunities if centered.
- Cp or Cpk = 1.33: A 4-sigma process. This is the most common minimum target for stable processes. It provides a buffer between the process spread and the spec limits.
- Cp or Cpk = 1.67: A 5-sigma process. Often required for safety-critical or significant characteristics, and for processes that show chronic instability.
- Cp or Cpk = 2.0: A 6-sigma process. World-class performance with extremely low defect rates.
Many automotive quality standards, including QS-9000-derived requirements, specify that stable processes should demonstrate a Cpk of at least 1.33. Processes that fall below this threshold typically require increased inspection until capability improves. For characteristics flagged as significant (safety or regulatory), a Cpk of 1.33 paired with a Ppk of 1.67 is a common requirement.
Cp/Cpk vs. Pp/Ppk
You’ll often see Pp and Ppk alongside Cp and Cpk. The formulas look identical, but they use different estimates of standard deviation, which changes what they tell you.
Cp and Cpk use within-subgroup variation only. This captures the short-term, inherent variability of the process when it’s running under consistent conditions. Pp and Ppk use the overall standard deviation calculated from all the data, which includes both within-subgroup and between-subgroup variation: shifts between operators, batches, time of day, and other sources of long-term drift.
Because Pp and Ppk include more sources of variation, they produce more conservative (lower) values. If your Cpk is 0.60 but your Ppk is 0.56, the gap indicates that between-subgroup variation is contributing to poorer long-term performance. A large gap between Cpk and Ppk signals that your process behaves differently across shifts, days, or production runs, even if it looks fine within any single subgroup.
When you’re unsure whether your sample meets all the conditions for rational subgrouping, it’s safer to rely on Pp and Ppk. They give you a more honest picture of what your process actually delivers over time rather than what it could deliver under ideal conditions.
Common Pitfalls
The most frequent mistake is calculating capability on a process that isn’t stable. If your control chart shows out-of-control signals (points beyond control limits, runs, or trends), the standard deviation estimate is unreliable and your capability index is meaningless. Stabilize the process first, then calculate.
Sample size matters more than people expect. With fewer than 50 data points, your capability estimate has wide uncertainty. A Cpk of 1.4 from 25 samples could easily reflect a true capability below 1.33. Larger samples give you confidence that the number you calculated is close to reality.
Non-normal data is the other trap. Capability indices assume a normal distribution. If your measurements are skewed or bounded (like flatness values that can’t go below zero), the standard formulas will overestimate or underestimate capability. Check a histogram or run a normality test before reporting. For non-normal data, percentile-based methods that mimic the ±3 standard deviation coverage of a normal distribution are a valid alternative.