Cpk vs Ppk: What's the Difference and When to Use Each

Why Cpk vs Ppk Trips Up Even Experienced Quality Engineers

Walk into almost any PPAP review and you will see the same confusion. The customer asks for Cpk. The supplier reports Ppk. Someone flips open a spreadsheet or pastes the measurements into a quick Cpk calculator, runs the numbers, and the two values look almost identical. Nobody is quite sure which one the AIAG manual actually wanted, so both get attached to the submission and everyone moves on.

It's a small issue until it isn't. Reporting the wrong capability index can mean rejecting a perfectly good process, accepting a drifting one, or failing a supplier audit. The difference between Cpk and Ppk is small in math but big in meaning, and once you see it clearly you will never mix them up again.

This guide explains what each index actually measures, how to calculate them correctly, when to use which one, and the common mistakes that make capability studies misleading.

A Quick Refresher on Process Capability

Process capability is a comparison. On one side you have how much variability your process produces. On the other side you have how much variability your customer is willing to accept. The capability index tells you the ratio.

Two things matter:

• Specification limits. The Upper Spec Limit (USL) and Lower Spec Limit (LSL) define what the customer calls "good." A shaft diameter of 10.00 mm plus or minus 0.05 mm gives you an LSL of 9.95 and a USL of 10.05.
• Process spread. This is how much the process output actually varies in production. It is usually expressed as 6 standard deviations because, for a normally distributed process, about 99.73 percent of output falls inside that range.

A process is "capable" when its spread fits comfortably inside the spec limits. Process capability indices are just different ways of measuring how comfortable that fit is.

The Four Indices at a Glance

The formulas look almost identical. The only difference is the sigma in the denominator. That one detail is what separates Cpk from Ppk, and it is the reason they can report very different values for the same data set.

What Cp and Cpk Measure

Cp and Cpk are short-term capability indices. They answer the question: "If the process were running under tight control and nothing changed, how capable would it be?"

To get them honestly, you need subgrouped data. A typical capability study collects 25 to 30 subgroups of 3 to 5 consecutive parts. The variation within each subgroup is assumed to be pure common-cause variation. That within-subgroup variation is used to estimate the standard deviation, usually written as sigma_within or sigma_ST (short-term).

Sigma_within is typically calculated from the average range of the subgroups (R-bar / d2) or the average subgroup standard deviation (S-bar / c4). These are estimates of how much the process varies when only inherent process noise is present.

Cp is the ratio of the specification width to the 6-sigma process spread, assuming the process is perfectly centered on the target:

Cp = (USL - LSL) / (6 * sigma_within)

Cpk corrects Cp for centering. A process can have a great Cp but still produce scrap if its mean has drifted toward one of the spec limits. Cpk uses whichever side of the distribution is closest to the nearest spec limit:

Cpk = min[(USL - mean) / (3 * sigma_within), (mean - LSL) / (3 * sigma_within)]

If Cp and Cpk are equal, the process is centered. If Cpk is much lower than Cp, the process is shifted off-target even if its spread is fine.

What Pp and Ppk Measure

Pp and Ppk are long-term performance indices. They answer a different question: "Looking at everything that happened across this entire production run, how well did the process actually perform?"

Instead of using within-subgroup variation, Pp and Ppk use the overall standard deviation of all the data points combined. This is the classic sample standard deviation you would compute in Excel with STDEV.S. In the literature it's called sigma_overall, sigma_total, or sigma_LT (long-term).

Pp = (USL - LSL) / (6 * sigma_overall)

Ppk = min[(USL - mean) / (3 * sigma_overall), (mean - LSL) / (3 * sigma_overall)]

The overall standard deviation captures every source of variation that occurred during the study: common cause noise plus tool wear, shift changes, material lot differences, temperature drift, and anything else that moved the process between subgroups.

This is why Ppk is almost always lower than Cpk. The overall sigma is larger than the within-subgroup sigma whenever there is any between-group variation, which there almost always is.

The Core Difference in One Sentence

Cpk uses within-subgroup sigma. Ppk uses overall sigma. That's it.

Everything else follows from that single choice:

• Cpk estimates what the process could do if you held it stable.
• Ppk measures what the process actually did over the time window you sampled.
• Cpk is typically higher than Ppk because within-subgroup sigma is smaller than overall sigma.
• The gap between Cpk and Ppk is a rough indicator of how much the process drifted or shifted during the study. Big gap means big drift.

A Worked Example

Say you are producing a shaft with a diameter spec of 10.00 plus or minus 0.05 mm. You collect 25 subgroups of 5 parts each over a full production shift. After running the math:

• Mean = 10.002 mm
• Within-subgroup sigma = 0.010 mm
• Overall sigma = 0.015 mm
• USL = 10.05, LSL = 9.95

Working through the indices:

Cp = (10.05 - 9.95) / (6 * 0.010) = 0.10 / 0.060 = 1.67

Cpk = min[(10.05 - 10.002) / 0.030, (10.002 - 9.95) / 0.030] = min[1.60, 1.73] = 1.60

Pp = (10.05 - 9.95) / (6 * 0.015) = 0.10 / 0.090 = 1.11

Ppk = min[(10.05 - 10.002) / 0.045, (10.002 - 9.95) / 0.045] = min[1.07, 1.16] = 1.07

Two important reads of this result:

1. Cpk of 1.60 means the process is very capable when it is running stable. The within-subgroup noise leaves plenty of room inside the spec limits.
2. Ppk of 1.07 means the process is barely capable over the full run. Something, maybe tool wear, maybe operator changeover, is adding enough between-subgroup variation to almost eat up the margin.

This kind of Cpk-Ppk gap is a red flag that deserves investigation before the process goes into full production.

When to Report Cpk vs Ppk

This is where the AIAG SPC manual and most customer requirements are very specific, and it's worth knowing the convention.

In automotive PPAP submissions, Ppk is the standard initial-capability index because the process is new and stability has not been proven. Once the process has been running stably for a while, Cpk takes over as the ongoing metric on the control charts.

Capability Thresholds and What They Mean

Most industries use the same rough thresholds for both Cpk and Ppk:

Automotive customers typically require Ppk of at least 1.67 for new processes and Cpk of at least 1.33 on ongoing production. Medical device and aerospace often demand 1.67 or higher on both. Consumer electronics tends to live in the 1.33 to 1.67 band.

A common mistake is treating these thresholds as pass-fail gates. A Cpk of 1.32 is not materially different from 1.33. What matters more is the trend over time and the gap between Cpk and Ppk.

Common Mistakes with Cpk and Ppk

1. Using sample standard deviation to compute Cpk. If you pull the STDEV.S function in Excel and plug it into the Cpk formula, you are really computing Ppk and calling it Cpk. This is the single most common error in capability reporting.

2. Computing capability on non-normal data. Both Cpk and Ppk assume the process output is approximately normally distributed. For one-sided-bounded data (flatness, runout, surface finish), for truncated distributions, or for heavily skewed data, a straight Cpk or Ppk value is misleading. Use a normality test first, and if it fails, consider a transformation or non-normal capability method like Weibull or Johnson.

3. Running a capability study on an unstable process. Capability indices only mean something if the process is in statistical control. A process with special causes bouncing around the chart can produce a nice looking Cpk that falls apart the moment you look away.

4. Too little data. Anything under about 100 data points gives you a wide confidence interval on your capability estimate. Customers often require 125 or more for PPAP for exactly this reason.

5. Reporting capability without the mean and sigma. A single Cpk number hides information. Always include the process mean, standard deviation, sample size, and the spec limits used. Without them, the index is unverifiable.

Calculating Cpk and Ppk in Practice

For a daily capability check, you have three reasonable options:

• Minitab or JMP. Industry standard, does everything correctly, costs real money.
• Excel with a good template. Accurate if the formulas are right, but error-prone when a colleague edits the wrong cell.
• A quick web calculator. Useful for fast checks, supplier data review, or validating a spreadsheet someone sent you.

For that third case we built a free Cpk calculator that takes raw measurements and spec limits and returns the mean, standard deviation, and all four capability indices (Cp, Cpk, Pp, Ppk) with a capability verdict. No signup, nothing to install. It runs entirely in the browser, which also means your data never leaves your machine. Good for running a sanity check on a supplier submission or double-checking Minitab's output before you send it on to a customer.

For a formal PPAP study, though, stick with your validated statistical software. A web calculator is for quick answers, not for documentation.

What a Good Capability Report Contains

Whether you use Minitab or something simpler, a capability report worth signing should include:

• Spec limits (USL, LSL, target)
• Sample size and sampling plan (subgroup size, frequency, duration)
• Mean and standard deviation (both within-subgroup and overall)
• Cp, Cpk, Pp, Ppk (all four, not just one)
• Histogram with spec limits overlaid
• Probability plot or normality test result
• Control chart showing process stability during the study
• Identification of any special causes observed

Reports that contain only a single Cpk number are almost always hiding something. Reports that show all four indices, the sigma values behind them, and a stability check are the ones auditors actually trust.

The Bottom Line

Cpk and Ppk answer different questions. Cpk is about what the process can do when it is behaving. Ppk is about what the process actually did across the whole study. Both are useful. Neither one tells the full story by itself.

For new processes and PPAP submissions, lead with Ppk. For ongoing SPC on a stable process, Cpk is the right ongoing metric. When the two disagree by a lot, don't paper over the gap. Investigate it, because it is telling you that something between your subgroups is adding variation the process wasn't supposed to have.

Get the sigma right, get the stability right, and both indices will tell you the truth.

---

Need to run a quick capability check? Try the free Cpk calculator from Vantage 8D. It returns Cp, Cpk, Pp, Ppk, and a capability verdict in seconds, with no signup required.