Cpk vs Ppk — or Why a Nice Number Doesn't Always Mean a Good Process

"Boss, we've got Cpk 1.67. We're good to produce."

I hear this sentence on the shop floor regularly. And it's not always a good basis for a decision.

Cpk only tells half the story. The other half — Ppk — can be a very unpleasant surprise. Especially when a customer complaint lands three months later, completely ignoring the nice number from PPAP.

This article is for quality engineers who know this and want to put it in order. And even more for those who have to work with Cpk and Ppk — project managers, production leaders, logistics heads — without being statisticians.

First, real life — what is "capability" actually?

Imagine your commute to work takes 30 minutes. That's your target. Your aim.

Sometimes you arrive in 27. Sometimes 34. Occasionally, when it's raining or there's a traffic jam, 45. And last week you got there in 22 minutes because it was a holiday morning and the road was yours.

Now your boss says: "I want you to arrive consistently in the 28-32 minute range." A specification. An upper and lower limit.

At that moment two questions appear:

• How evenly do you hit that range when it's going well?
• How often do you actually fit inside it, across all the days of the month?

The first question is about your capability. The second is about your performance.

Cpk and Ppk are two mathematical ways to ask these two questions in manufacturing. Except — each looks at a different window of your data.

Cpk — the short view, the ideal day

Cpk is the process capability index. It takes data from a short window. From a single shift, a single batch, a single section of production. The best equipment, the most focused operator, the cleanest raw material.

In the commute analogy: it's your time on a Wednesday at 8:30 AM, sunny weather, no accidents, no snowplow. Best-case scenario.

Cpk says: under stable conditions, you can hit 28-32 minutes consistently.

Mathematically it uses the short-term standard deviation — variability within a subgroup. The assumption is that the process is in statistical control. No special causes are interfering.

Cpk ≥ 1.33 means: the process is capable. Under ideal conditions.

Ppk — the long view, the whole month

Ppk is the process performance index. It takes data from the entire period. From all shifts, all batches, all days. Monday morning and Friday afternoon. Summer morning and winter rain.

In the commute analogy: it's every single trip you made to work in the last month. Including that Wednesday in 22 minutes and that Friday in 47.

Ppk says: how many of your actual trips really fit inside the 28-32 range.

Mathematically it uses the overall standard deviation — variability between and within subgroups. It includes every special cause that happened during the period.

Ppk ≥ 1.33 means: the process produces in spec in real life. Not just under ideal conditions.

Here's where it gets interesting

Cpk and Ppk are almost never equal. And when they're significantly different, that's a signal.

Picture this:

Your Cpk is 1.67. As if you arrived in 28-32 minutes every Wednesday morning, ten times in a row across two weeks. Beautiful numbers, pretty statistics.

But your Ppk is 0.75. Because across the whole month your data also includes those Mondays at 45, Fridays at 47, Thursdays when you drove a friend to work.

Cpk says: you've got it down. Ppk says: you arrive late 30 % of the time.

Both are true. They differ in the time window.

Why this drives quality folks crazy in automotive

Because the gap between "a shift when everything is fine" and "a quarter running 24/7" is enormous. The customer doesn't count your Cpk from Friday, April 11th. The customer receives parts every day, in every weather, from every shift.

When an OEM (Volkswagen, Stellantis, BMW, Toyota — pick one) asks for Cpk ≥ 1.33 in a PPAP, it often means: I want assurance that you can produce in spec. Capability.

When they ask for Ppk ≥ 1.33, they're saying something else: I want you to actually produce in spec. Performance. Real-world output.

And this is exactly where the typical automotive pitfall is born: a quality engineer shows a pretty Cpk in PPAP because the data was collected from a single controlled batch under perfect conditions. The boss sees 1.67 and reports it to the customer. PPAP gets approved.

Three months later, a complaint arrives. Because the Ppk across the real three months of production was 0.89.

And nobody looked at that during PPAP.

Common mistakes I've seen

• Confusing Cpk and Ppk in the report. Sometimes deliberately, sometimes unknowingly. It always comes out — at the latest during the customer complaint.
• Treating Cpk from 30 parts as representative for the whole production. 30 parts is a short-term snapshot. Not the whole story.
• Submitting PPAP with Cpk 1.67 and being shocked six months later that production has a scrap rate as if Cpk were 0.9. PPAP Cpk is not a prediction of the future. It's a photograph of one moment.
• Calculating Cpk on data that includes calibration drift. A few points from a single afternoon after an operator change can shoot your Cpk up — but it tells you nothing about reality.
• Not collecting Ppk in parallel with Cpk. This is the quiet mistake. If you never look at Ppk, you have no way to know that special causes are slipping through your process.

When to use which

Use Cpk when you want to answer: "Is the process capable of producing in spec under stable conditions?"

That's useful for:

• process development and ramp-up (before SOP),
• part of PPAP,
• statistical process control (SPC),
• daily on-line checks.

Use Ppk when you want to answer: "Am I actually producing in spec under real conditions?"

That's useful for:

• customer scorecards,
• complaint analysis,
• deciding whether to release a process into full production,
• monthly quality reviews for management.

And when you have both, compare them. This is the most valuable piece of information most quality engineers overlook:

• Cpk higher than Ppk → the process is capable, but it doesn't hold up in the long run. Special causes are slipping through somewhere — shift, material, calibration, people, environment.
• Cpk ≈ Ppk → the process is stable. Short-term and long-term performance are similar. A good signal.
• Cpk lower than Ppk → mathematically rare. It usually indicates a data collection error or incorrect subgrouping.

The bottom line

When someone shows you Cpk and proposes a decision, ask one single question:

"And what's the Ppk?"

If they don't know or don't have it, don't wait three months for the complaint to arrive. Ask again. Keep asking until you get it.

Capability and performance are not the same thing. Ability and output are not the same thing. And in automotive, that difference tends to be expensive.