Say your customer's PPAP checklist asks for "Cpk 1.33 minimum", and your capability report shows a Cpk of 1.45 and a Ppk of 1.12 (illustrative numbers). Is the part approved or not? The answer depends on knowing what each number measures, and why they can differ so much on the same data.

What process capability compares

The NIST/SEMATECH e-Handbook puts it simply: process capability compares the output of an in-control process to the specification limits using capability indices. Two ideas sit inside that sentence:

  • The spec limits come from the drawing: the upper limit (USL) and lower limit (LSL) of a dimension.
  • The process spread comes from your measurements: the mean and the standard deviation.

A capability index tells you how many times the process spread fits inside the tolerance band, and how well centred it is.

Cp and Cpk: can the process fit, and is it centred?

NIST gives the basic formulas:

  • Cp = (USL - LSL) / 6σ. This only asks whether the spread fits inside the tolerance. It ignores where the process is centred.
  • Cpk = the smaller of (USL - mean) / 3σ and (mean - LSL) / 3σ. This checks the distance from the mean to the nearer limit, so an off-centre process scores lower.

Because Cpk looks at the worse side, Cpk is never more than Cp. When the two are close, the process is well centred. When Cpk is much lower than Cp, the spread is fine but the setting is off, which is often a simple offset correction on the machine.

NIST's own worked example has USL 20, LSL 8 and a standard deviation of 2. Cp comes out at 1.0, but Cpk is only 0.67 because the mean sits close to the upper limit. NIST calls it "not a good process".

Ppk: the same formula with a different sigma

This is where most confusion comes from. As the Minitab blog on Cpk vs Ppk explains, the formulas are nearly identical, and the only difference lies in the denominator:

  • Cpk uses the within standard deviation, the variation inside each subgroup. If you measure 5 parts every hour, each hourly group of 5 is a subgroup.
  • Ppk uses the overall standard deviation, the variation of all the data taken together.

So Cpk does not see the shift and drift between subgroups: a tool wearing through the shift, a warm-up after lunch, a new bar lot, a different setter. Minitab describes Cpk as the potential capability, and Ppk as "where you are at the end of the proverbial day".

Minitab's capability statistics guide adds that within variation is due to common causes only, while overall variation includes special causes as well.

A high Cpk with a low Ppk is not a contradiction: it is your process telling you that something changes between subgroups.

Reading the two together

  • Cpk and Ppk about equal: Minitab notes that a stable process with little variation over time gives two statistics that are about the same. This is what you want.
  • Cpk much higher than Ppk: the machine can hold the tolerance in the short term, but something shifts over time. Look at tool wear compensation, temperature, material lots and setup changes.
  • Both low: the basic spread is too wide for the tolerance. Look at the process route, fixture rigidity, or the gauge itself.

What values do customers ask for?

Minitab notes that many industries use 1.33 as the benchmark for both Cpk and Ppk. Requirements differ between customers, so always read the customer-specific requirements or the supplier quality manual for the exact numbers and which index they mean. Minitab also points out that some customers use the word "Cpk" for what Minitab calls Ppk, so confirm which formula they expect.

NIST's table shows why the numbers matter. For a centred process, a Cp of 1.00 means about 0.27% out of tolerance, 1.33 means about 64 parts per million, and 1.66 about 0.6 parts per million.

Practical tips for a capability study

  • Collect enough data. NIST says capability estimates are generally valid with about 50 independent values, and recommends 100 or more for capability studies.
  • Use rational subgroups. Take consecutive parts made under the same conditions, for example 5 parts every hour.
  • Check stability first. Capability indices assume an in-control process, so plot a control chart before you report a Cpk.
  • Check normality. Cp and Cpk assume normally distributed data. Features like flatness or runout, which cannot go below zero, often are not normal.

For example (illustrative numbers): a bore of 20.000 to 20.021 mm with a mean of 20.012 mm and a within standard deviation of 0.002 mm gives a Cpk of (20.021 - 20.012) / 0.006 = 1.5. If the overall standard deviation across the full shift is 0.003 mm, Ppk drops to 1.0, below a 1.33 requirement.

How SpecReview helps

Capability studies start with knowing which characteristics matter. SpecReview reads your drawing PDF or STEP model and builds a quality plan, a layout inspection plan and a PPAP document set, so the characteristics that need capability evidence are listed before the sample run. Its DFM and rejection risk list also flags tight tolerances early, with a short reason why each matters.