It depends on the industry and the customer specification. 1.33 is a commonly used minimum; some fields ask for more. Below 1 is generally considered inadequate. Take the required value from your customer specification.

What if Cp is high but Cpk is low?

The process is narrow but off-centre. The problem is the position of the mean, not the variation; an adjustment or calibration often fixes it.

Why do Cpk and Ppk differ?

Cpk uses short-term, Ppk overall variation. If Ppk is clearly lower, the process is drifting over time or has special causes; check stability with a control chart.

How many measurements are needed?

At least 30 measurements are recommended. With less data the indices have a wide confidence interval; the tool gives a result but shows a warning.

What happens with a one-sided specification?

If you enter only a USL or only an LSL, only Cpk and Ppk are calculated; Cp and Pp need two limits and are undefined. A surface roughness with only an upper limit is evaluated this way, for example.

Do the measurements have to be in time order?

For Cp/Cpk yes: short-term variation is estimated from the differences between consecutive measurements. If the order is scrambled Cp/Cpk lose their meaning. Pp/Ppk and the ppm (with overall s) do not depend on the order.

Cp / Cpk process capability calculator

Enter your measurements and specification limits. See Cp, Cpk, Pp and Ppk, the expected defect rate (ppm) assuming a normal distribution and a histogram.

Free tool · Predictive maintenance

Measurements (in time order)

One value per measurement; separator space, new line, tab or ; (a column from Excel can be pasted). Use a point or a comma for decimals. Cp/Cpk estimates short-term variation from consecutive measurements, so values must be in production order. At least 30 values are recommended.

Upper specification limit (USL) (optional)If left empty only the lower limit is evaluated.Lower specification limit (LSL) (optional)If left empty only the upper limit is evaluated.

The calculation runs in your browser; the measurements you paste are not sent anywhere.

40 values read.

Cpk (short term, includes centring): 1.17
Cp (short term, spread): 1.26
Ppk (long term, includes centring): 1.23
Pp (long term, spread): 1.33
Mean: 10.007
Short-term sigma σ̂ = MR̄ / 1.128: 0.026368
Overall standard deviation s: 0.024987
Number of values: 40
Measurements outside specification: 0 (0%)
Skewness (0 for normal): -0.38
Excess kurtosis (0 for normal): 0

Assessment (Cpk)

Cpk between 1 and 1.33: marginal; small shifts can lead to out-of-specification output.

The required value depends on the industry, the customer specification and how critical the product is (some fields ask for more); take the thresholds from your own specification.

Histogram and specification limits

Normal curve (overall s)
Bar outside specification

Expected ppm under normality

Expected ppm under normality
	With short-term σ̂	With overall s
Below the lower limit	22.8	8.46
Above the upper limit	226	107
Total	249	115

ppm = out-of-specification parts per million. The estimate assumes the data is normally distributed and is very sensitive in the tails; the observed count and the real defect rate can differ.

Cp/Cpk use short-term variation (between neighbouring measurements), Pp/Ppk the overall standard deviation of all the data. A clear gap between Cpk and Ppk means the process is drifting over time or is not stable; confirm stability with a control chart first.

The calculations assume a normal distribution. If the data is clearly skewed or very peaked, the indices and the ppm estimate can be misleading.

Cp = (USL − LSL) / 6σ̂ · Cpk = min(USL − X̄, X̄ − LSL) / 3σ̂ · σ̂ = MR̄ / 1.128 · Pp and Ppk use the same formulas with overall s · ppm = 10⁶ · [Φ((LSL − X̄)/σ) + Φ(−(USL − X̄)/σ)]

Capability indices are meaningful only for a stable (in control) and approximately normally distributed process; little data increases the uncertainty. This tool gives a first assessment; for a formal capability study apply the method required by the customer specification.

Let's look at why capability drops in your process data and which deviations could be spotted in advance.

Request a conversation

How to use it

A
Paste your measurements in production order or load the example data.
B
Enter the specification limits (USL, LSL; one is enough for a one-sided specification).
C
Read Cp/Cpk, Pp/Ppk, the ppm estimate and the histogram; pay attention to the gap between Cpk and Ppk.

Cp and Cpk: spread and centring

Cp is the ratio of the specification width to the process spread: (USL − LSL) / 6σ. If the process mean sits exactly midway between the limits, Cp tells you whether the spread fits. But it does not know where the mean is: a narrow process running close to a limit has a high Cp and a low Cpk.

Cpk also accounts for centring: the distance from the mean to the nearest specification limit divided by 3σ. Cpk is always equal to or smaller than Cp; if they are equal the process is perfectly centred. A Cpk of 1 means the nearest limit is 3σ from the mean.

Pp/Ppk versus Cp/Cpk

For Cp/Cpk, σ is estimated from short-term variation (within subgroups, here between consecutive measurements); it shows what the process can do when stable. For Pp/Ppk, σ is the overall standard deviation of all the data; it shows what actually happened, including drift and fluctuation.

If the process is stable the two are close. If Ppk is clearly lower than Cpk there is a drift or a special cause over time: first establish stability with a control chart, then assess capability.

The ppm estimate and the normality assumption

ppm is the expected number of out-of-specification parts per million; it is computed from the tail area of the normal curve beyond the specification limits. The tool calculates the standard normal distribution function directly (relative accuracy in the tail about 10⁻¹³).

The estimate itself is limited: if the data is not normal, especially in the tails, the result can be many times too low or too high. Skewness and kurtosis give an idea (both are 0 for a normal distribution) but are not a normality test. With small samples the ppm estimate is not reliable; read it together with the observed count of out-of-specification measurements.

FAQ

What should Cpk be?: It depends on the industry and the customer specification. 1.33 is a commonly used minimum; some fields ask for more. Below 1 is generally considered inadequate. Take the required value from your customer specification.
What if Cp is high but Cpk is low?: The process is narrow but off-centre. The problem is the position of the mean, not the variation; an adjustment or calibration often fixes it.
Why do Cpk and Ppk differ?: Cpk uses short-term, Ppk overall variation. If Ppk is clearly lower, the process is drifting over time or has special causes; check stability with a control chart.
How many measurements are needed?: At least 30 measurements are recommended. With less data the indices have a wide confidence interval; the tool gives a result but shows a warning.
What happens with a one-sided specification?: If you enter only a USL or only an LSL, only Cpk and Ppk are calculated; Cp and Pp need two limits and are undefined. A surface roughness with only an upper limit is evaluated this way, for example.
Do the measurements have to be in time order?: For Cp/Cpk yes: short-term variation is estimated from the differences between consecutive measurements. If the order is scrambled Cp/Cpk lose their meaning. Pp/Ppk and the ppm (with overall s) do not depend on the order.

Raise capability with data

Let's look at which deviations in your machine and process data pull capability down, and whether they could be seen in advance.

Request a conversation Explore the predictive maintenance service →