You pull data from a control chart, run the numbers, and get two different answers depending on whether you calculate Cpk or Ppk. That gap confuses even experienced quality engineers, and it matters because auditors and customers will ask you to explain it. Process capability vs process performance comes down to one variable: which standard deviation you use, and that choice changes what the number actually tells you about your process.
Process capability (Cp/Cpk) measures your process using within-subgroup variation, the short-term, inherent variation you’d see if nothing else changed. Process performance (Pp/Ppk) uses the overall standard deviation, capturing everything that happened across your entire data set, including shifts, drift, and tool wear between subgroups. Capability answers "what can this process do under controlled conditions?" Performance answers "what did this process actually deliver?"
In this article, we’ll walk through the exact formulas for each index, show you how subgroup size affects the calculations, and give you clear rules for when to report Cpk versus Ppk on a PPAP submission or supplier audit. You’ll also see a worked example using real sample data, so you can apply this the next time your numbers don’t match.
Why the capability vs performance distinction matters
Confusing Cpk with Ppk isn’t just a technicality, it can get a supplier disqualified or a part rejected during a PPAP submission. The two indices answer different questions, and if you report the wrong one, you’re either overstating what your process can reliably do or understating how well it’s actually performing. This distinction shows up constantly in automotive, aerospace, and medical device manufacturing, where quality frameworks explicitly require both numbers at different stages of production.
The math tells two different stories
Cpk relies on the pooled standard deviation calculated from variation within each subgroup, essentially averaging out the noise that happens between samples taken close together in time. Ppk uses the total standard deviation of every individual data point in your dataset, so it picks up drift, tool wear, operator changes, and material lot variation that Cpk deliberately filters out. When your process is stable and in control, these two numbers converge and tell nearly the same story. When your process has shifts or trends, the gap between them widens, and that gap itself becomes diagnostic information.
A big gap between Cpk and Ppk almost always means your process has special-cause variation hiding between subgroups, not a calculation error.
Getting it wrong costs real money
Reporting Cpk when a customer asked for Ppk, or the other way around, isn’t a paperwork slip. Quality engineers who submit an inflated Cpk on a new part introduction are effectively promising a capability the process hasn’t demonstrated yet under real production conditions. Auditors trained on standard PPAP documentation requirements know to check which formula generated your number, and a mismatch raises questions about whether your team understands its own data, let alone whether the part is production-ready.
Standards and customer requirements assume you know the difference
Automotive OEMs and their suppliers typically expect Ppk during initial production runs, since you haven’t accumulated enough long-term data to calculate a trustworthy Cpk yet. Once the process has run long enough to establish stability, usually 20 to 25 subgroups collected over multiple shifts or days, you shift to reporting Cpk as your ongoing capability metric. Skipping this progression, or reporting Cpk from a single short run, gives everyone a false sense of security about a process that hasn’t proven it can hold up over time.
Here’s a quick reference for when each mismatch tends to cause problems:
| Situation | What goes wrong |
|---|---|
| Reporting Cpk during initial qualification | Overstates capability before long-term data exists |
| Reporting Ppk for ongoing production control | Understates capability by including known transient issues |
| Ignoring a large Cpk-Ppk gap | Misses signal of special-cause variation between subgroups |
| Using the wrong subgroup size in the Cpk formula | Distorts the within-subgroup variance estimate |
The distinction protects your improvement priorities
Teams chasing a single capability number without knowing which one they’re looking at end up solving the wrong problem. If Ppk is low but Cpk is high, your process is capable in short bursts, so the real issue is stability rather than the process itself, and you should hunt for assignable causes like shift changes or incoming material variation. If both indices are low together, the process lacks the inherent capability to meet spec, and no amount of stabilization work will fix it, you need a design or equipment change instead. Getting this diagnosis right up front saves weeks of chasing the wrong root cause.
How to calculate and interpret Cp, Cpk, Pp, and Ppk
Each of these four indices follows a similar shape, but the standard deviation you plug in changes everything downstream. Cp and Pp measure raw spread against your tolerance width, while Cpk and Ppk fold in how far your process mean sits from the nearest spec limit. Six Sigma practitioners often memorize the formulas without understanding why the denominator changes, and that’s exactly the gap that causes reporting errors later.
The four formulas side by side
Here’s how the calculations line up once you separate spread-only indices from centering-adjusted ones:
| Index | Formula | Standard deviation used |
|---|---|---|
| Cp | (USL – LSL) / 6σ_within | Pooled within-subgroup σ |
| Cpk | min[(USL – X̄)/3σ_within, (X̄ – LSL)/3σ_within] | Pooled within-subgroup σ |
| Pp | (USL – LSL) / 6σ_overall | Overall sample σ |
| Ppk | min[(USL – X̄)/3σ_overall, (X̄ – LSL)/3σ_overall] | Overall sample σ |
USL and LSL are your upper and lower spec limits, and X̄ is your grand average across all the data.
Reading Cp and Pp: the spread check
Getting a Cp or Pp value above 1.33 tells you the tolerance band is comfortably wider than your process spread, which is the baseline most manufacturers target for a stable production process. Neither index cares where your process mean sits relative to spec, so a Cp of 2.0 with a mean sitting right on the upper spec limit still means you’re shipping defects. That’s the trap: teams report Cp or Pp alone and assume a high number equals a healthy process, when it only confirms the spread is narrow enough, not that the parts are centered.
Reading Cpk and Ppk: the centering check
Interpreting Cpk and Ppk requires looking at both halves of the min[] formula, because whichever spec limit is closer to your mean determines the final number. A Cpk of 1.0 typically corresponds to a process producing roughly 2,700 defects per million opportunities if the distribution is normal, and that math scales quickly as the index drops.
If Cpk or Ppk is meaningfully lower than Cp or Pp, your process is capable but off-center, and shifting the mean fixes more defects than reducing variation ever will.
Just remember that Ppk will almost always run slightly lower than Cpk on the same dataset, because it’s absorbing the between-subgroup noise that Cpk’s pooled calculation strips out.
A quick calculation walkthrough
Quality software will do the arithmetic for you, but knowing the sequence helps you catch a mistake before it lands in a customer report.
- Collect subgroups (typically 4-5 pieces per subgroup, 20+ subgroups total).
- Calculate the pooled within-subgroup σ for Cp/Cpk, and the overall σ across every individual point for Pp/Ppk.
- Compute X̄ as the grand mean of all data.
- Plug both standard deviations into the four formulas above.
- Compare Cpk to Ppk directly. A gap wider than about 0.2 to 0.3 usually signals instability worth investigating before you trust either number.
When to use process capability versus process performance
Deciding between Cpk and Ppk starts with one question: has your process proven it’s stable over time? New processes and new parts haven’t earned the right to a Cpk claim yet, because Cpk assumes you already know what "normal" variation looks like within a subgroup. Ppk makes no such assumption, so it’s the honest choice whenever you’re looking at a limited data set from a process that hasn’t run long enough to show its long-term behavior.
Early production and PPAP submissions
For a new part introduction, a first production run, or a PPAP submission, report Ppk. You typically have somewhere between one and three production runs’ worth of data, not the weeks of history needed to trust a pooled within-subgroup estimate. Automotive OEMs following AIAG guidance expect Ppk during this phase precisely because it reflects everything that happened, shift changes, material lot swaps, tooling break-in, without pretending the process has already settled into a repeatable pattern.
Established, stable production runs
Once your control charts show sustained statistical control, usually after 20 to 25 subgroups collected across multiple shifts, switch to Cpk as your ongoing metric. At this point Cpk gives you a cleaner read on what the process is inherently capable of, stripped of the between-subgroup noise that Ppk still carries. Ongoing capability studies, annual requalification, and internal process monitoring should lean on Cpk because you’re now managing a known, characterized process rather than proving one for the first time.
Use Ppk when you’re proving a process for the first time, and Cpk once you’re managing a process you already trust.
A simple decision checklist
Run through this before you decide which index to report:
- Data history under 25 subgroups? Report Ppk.
- Control charts show no out-of-control points across 20+ subgroups? Report Cpk as your primary metric, but keep Ppk as a cross-check.
- Customer contract or PPAP form specifies one explicitly? Follow that requirement regardless of what your internal preference is.
- Large, unexplained gap between Cpk and Ppk? Don’t report either with confidence yet; investigate the special-cause variation first.
- Multi-site production of the same part? Report both, since site-to-site drift often only shows up in Ppk.
Technically minded auditors will ask which formula produced your number, so knowing this checklist cold saves you from an awkward conversation during a supplier audit. Situations involving multi-site organizations deserve extra caution here, since a part that looks capable at one plant based on Cpk alone might be masking real variation that only shows up when you calculate Ppk across combined data from every location. Reporting both numbers side by side, rather than picking one and hiding the other, is what separates a mature quality system from one that’s just checking a box.
Examples that show capability and performance in action
Numbers on a spec sheet only click once you see them applied to an actual part. Below are two worked examples pulled from typical shop-floor scenarios: a stable turning operation and a injection molding process fighting tool wear. Both use the same spec width, so you can see exactly how the gap between Cpk and Ppk shifts depending on what’s happening between subgroups.
A stable process where the numbers agree
Picture a shaft diameter with a spec of 9.50mm to 10.50mm, and a grand mean of 10.02mm across 25 subgroups of five pieces each. The pooled within-subgroup sigma comes out to 0.15mm, while the overall sigma across every individual reading is 0.16mm, barely different because the process hasn’t drifted shift to shift.
| Metric | Formula result | Value |
|---|---|---|
| Cpk | min(0.48/0.45, 0.52/0.45) | 1.07 |
| Ppk | min(0.48/0.48, 0.52/0.48) | 1.00 |
That 0.07 gap is exactly what you want to see: close enough that a customer reviewing your PPAP package trusts both numbers tell the same story. Stability like this is the reward for a well-maintained machine and consistent operators, not luck.
A drifting process where the gap tells the real story
Now take an injection molding operation holding a wall thickness spec of 9.50mm to 10.50mm, centered right on 10.00mm. Early in the shift, the tool runs tight, giving a within-subgroup sigma of just 0.08mm. But across three shifts, tool wear pushes individual readings wider apart, producing an overall sigma of 0.18mm once you look at every data point together.
| Metric | Formula result | Value |
|---|---|---|
| Cpk | min(0.50/0.24, 0.50/0.24) | 2.08 |
| Ppk | min(0.50/0.54, 0.50/0.54) | 0.93 |
A Cpk of 2.08 looks fantastic on paper, and it’s not wrong, the process genuinely runs tight within any given subgroup. But a Ppk of 0.93 tells the customer something Cpk hides completely: over the full production window, this process isn’t holding that tight window at all.
A Cpk near 2.0 paired with a Ppk under 1.0 isn’t a calculation error, it’s tool wear or drift hiding in plain sight.
What the two examples teach you
Comparing these side by side shows why reporting only one index is risky. The stable shaft example proves your process is genuinely in control, so either number earns customer trust. The molding example proves the opposite: without Ppk, you’d have shipped a PPAP submission claiming near-perfect capability on a process that’s actually drifting toward the spec limit by the third shift. Catching that gap early sends you looking at heater bands, cooling time, or mold maintenance instead of celebrating a Cpk number that never told the whole story.
Choosing the right metric for your process
The process capability vs process performance question isn’t academic once you’re staring at a supplier scorecard or a rejected PPAP submission. Cpk tells you what your process can do under controlled conditions, Ppk tells you what it actually did, and a mature quality system reports both instead of hiding the gap. Getting this right matters most when a customer, an auditor, or your own leadership is deciding whether to trust a number you calculated under pressure.
Getting comfortable with these formulas takes more than reading definitions, it takes practice pulling real data and defending your numbers to someone who knows the difference. If your team is still second-guessing which index belongs on a submission, or you need hands-on Six Sigma training that covers this exact skill, reach out to our team and we’ll help you build the capability, and the confidence, to back it up.