You can run a process that behaves perfectly, never drifts, and still ship parts the customer rejects. That gap, between a process that is stable and a process that actually meets specification, is what process capability measures. Cp and Cpk are the two numbers that put a figure on it, and knowing the difference between them is one of the most useful things a quality or manufacturing team can carry into a daily production meeting.
This guide explains Cp and Cpk in plain language, walks through a worked example you can copy, and shows how capability ties back to statistical process control and the quality side of your Overall Equipment Effectiveness (OEE).
Process capability compares the natural spread of your process to the width of the specification it has to fit inside. The specification is set by the customer or the design: a shaft must be 20.00 mm plus or minus 0.30 mm, for example. Your process, meanwhile, has its own spread, the normal variation it produces when nothing is wrong. Capability asks a simple question: how comfortably does the process spread fit inside the specification window?
If the process spread is much narrower than the specification, you have room to spare and few defects. If the spread is as wide as the specification, or wider, you will make out-of-spec parts even when nothing unusual happens. Cp and Cpk turn that comparison into a single number.
Cp measures how the width of your process compares to the width of the specification, assuming the process is perfectly centered between the two limits. The formula is Cp = (USL minus LSL) divided by six standard deviations, where USL and LSL are the upper and lower specification limits.
A Cp of 1.0 means the process spread exactly fills the specification. A Cp of 2.0 means the specification is twice as wide as the process needs, a very comfortable margin. The catch is that Cp ignores where the process is actually centered. It tells you what the process could do if it were perfectly aimed, not what it is doing right now.
Cpk accounts for centering. It measures the distance from the process average to the nearer specification limit, expressed in units of the process spread. Because it always takes the worse of the two sides, Cpk drops as soon as the process drifts off center, even if the spread itself stays the same. That is why Cpk is the number most customers ask for: it reflects reality, not potential.
A useful way to hold the two ideas together: Cp is the size of the car, Cpk is whether you have parked it in the middle of the garage. A small car badly parked still scrapes the wall.
Suppose a shaft diameter is specified at 20.00 mm with a tolerance of plus or minus 0.30 mm. That gives you:
You sample the line and find the process is centered at a mean of 20.10 mm with a standard deviation of 0.06 mm.
Cp = (20.30 minus 19.70) divided by (6 times 0.06) = 0.60 divided by 0.36 = 1.67. On spread alone, the process looks strong.
Cpk takes centering into account. The mean of 20.10 sits closer to the USL than to the LSL, so the upper side governs the result:
Cpk is the smaller of the two, so Cpk = 1.11. The gap between Cp at 1.67 and Cpk at 1.11 is the cost of running off center. Recenter the process at 20.00 mm and Cpk would rise to meet Cp at 1.67, with no change to the equipment and no reduction in spread. That single insight, that a lot of capability is lost to poor centering rather than excess variation, is often the fastest quality win available on the floor.
| Cpk | What it means | Approximate defect level |
|---|---|---|
| Below 1.00 | Not capable; producing out-of-spec parts | High |
| 1.00 | Barely meeting spec, no margin for drift | ~2,700 ppm |
| 1.33 | Common minimum for a capable process | ~63 ppm |
| 1.67 | Strong capability, typical automotive target | ~0.6 ppm |
| 2.00 | Six Sigma level capability | ~2 ppb |
Targets vary by industry and by how critical the characteristic is. Automotive suppliers often require a Cpk of 1.33 or 1.67 on key characteristics, while a non-critical dimension might be acceptable at 1.0. The point is not to chase a universal number but to agree the target with the customer and the risk involved.
Capability numbers are only trustworthy when the process is stable. You establish stability first with statistical process control and control charts, then measure capability. Running a capability study on an unstable process produces a number that means very little, because the spread and the average are both still moving. Get the process in control, confirm it, then calculate Cp and Cpk.
There is also a related pair, Pp and Ppk, known as process performance. They use the overall long-term variation rather than the short-term within-subgroup variation behind Cp and Cpk. Cp and Cpk describe short-term potential; Pp and Ppk describe how the process actually performed over a longer window. Many quality systems report both, and the gap between them is a useful measure of how much the process shifts over time.
Capability lives on the Quality side of OEE. Poor capability shows up as quality losses, the defects and reduced yield that are part of the Six Big Losses. A process running at a Cpk of 1.0 will quietly generate scrap and rework that pulls the Quality factor, and therefore OEE, down. Improving capability is one of the most direct ways to lift the Quality component without touching availability or speed.
Cp and Cpk are only as good as the measurements behind them. If data is sampled by hand, recorded at the end of a shift, or pulled from a gauge that is not properly calibrated, the standard deviation you calculate is wrong, and so is every capability number built on it. This is the same data-quality gap that undermines manual OEE tracking. When measurements come straight from the process in real time through automated data capture, a capability study stops being a once-a-quarter audit exercise and becomes a live signal you can act on.
It is also why this groundwork matters before any analytics or AI initiative. Clean, structured, trustworthy operational data is the foundation, and even the largest manufacturers have learned that you cannot model or predict your way out of messy measurements. Sort the data first, and capability, SPC, and everything downstream gets easier.
Cp measures only the spread of the process against the specification width and assumes the process is perfectly centered. Cpk also accounts for how far off center the process actually runs. Cp is the potential; Cpk is the reality. When a process is perfectly centered the two are equal, and the further off center it drifts, the more Cpk falls below Cp.
A Cpk of 1.33 is a common minimum for a capable process, and 1.67 is a typical target for critical automotive characteristics. A Cpk of 1.0 means you are just meeting specification with no margin for drift. The right target depends on the industry and how critical the characteristic is, so agree it with the customer.
Cpk uses short-term, within-subgroup variation and describes the potential capability of a stable process. Ppk uses the overall, long-term variation and describes how the process actually performed over a longer period. Cpk usually looks better than Ppk, and the gap between them reflects drift and shifts over time.
You can run the numbers, but they will not mean much. Capability assumes a stable, predictable process. If control charts show the process is out of control, stabilize it first, then measure capability.
Cp and Cpk only help when the measurement data behind them is accurate and current. Fabrico captures production and quality data straight from your machines in real time, keeps it clean and structured, and surfaces losses as they happen, the same foundation your SPC program and any future AI initiative depend on. Book a short demo to see how it maps to your lines, or start with the OEE basics.
Login
Blog
