Quick answer: Cpk is a risk score, not a grade. It tells you how many standard deviations sit between your process mean and the nearest specification limit. Cpk = 1.33 means four standard deviations of clearance and roughly 64 defective parts per million. Cpk = 1.0 means three standard deviations and roughly 2,700 ppm. Every drop in Cpk is a rise in the probability that your next shipment contains defects.
1. What Process Capability Actually Measures
Process capability is not the same thing as process control. A process can be perfectly stable (no trends, no special-cause signals, every point inside the control limits) and still produce defects. Stability tells you the process is predictable. Capability tells you whether that predictable output actually fits inside the tolerance your customer requires.
Think of it as a parking problem. A control chart tells you the driver is consistent: same approach angle, same speed, same parking position every time. Process capability tells you whether the car actually fits in the garage. A Mini driven consistently will fit every time. A transit van driven consistently will clip the walls every time, no matter how stable the approach.
The comparison works by forming a ratio. Specification width (USL minus LSL) goes in the numerator. Process width, measured as six standard deviations, goes in the denominator. The 6σ comes from the fact that 99.73% of values in a normal distribution fall within ±3σ of the mean. If the specification width is larger than the process width, the ratio exceeds 1.0 and the process is theoretically capable.
2. Cp vs Cpk: Potential vs. Reality
Cp measures what could happen. Cpk measures what is happening.
Cp compares the specification width to the process spread, assuming the process mean sits exactly at the midpoint of the specification range. It answers one question: if we got the centring perfect, would the process fit?
Cpk compares the distance from the process mean to the nearer specification limit, divided by three standard deviations. It answers a different question: right now, with the mean where it actually is, how much clearance do we have on the worst side?
Cp = (USL − LSL) / (6σ)
Cpk = min[(USL − μ) / (3σ), (μ − LSL) / (3σ)]
Cpk is always equal to or smaller than Cp for the same data. The gap between them is a diagnostic. Cp measures only spread. It does not measure whether the process is centred. Cpk measures both spread and centredness by looking at the worst-case side.
A large Cp with a small Cpk means the process is capable in principle but poorly centred. Fix the targeting, not the variation. A small Cp with an equally small Cpk means the spread itself is the problem. You need a different machine, a different material, or a different method.
3. What the Numbers Actually Mean: A Cpk-to-Risk Decoder
Every Cpk value corresponds to a specific defect rate, expressed in parts per million. The table below translates the numbers into plain risk.
| Cpk | Sigma Level | Defect Rate (ppm) | What It Means |
|---|---|---|---|
| Less than 0 | N/A | More than 500,000 | The process mean is outside the specification limits. You are making more bad parts than good ones. Stop the line. |
| 0.67 | 2.0 | ~45,500 | The process is producing defects at a rate of roughly one in twenty-two. Unsustainable. |
| 1.00 | 3.0 | ~2,700 | Barely capable. The 6σ spread of the process exactly fills the specification width. Any shift in the mean and you are shipping defects. |
| 1.33 | 4.0 | ~64 | Industry minimum for a capable process. The process mean sits four standard deviations from the nearest limit. You have a safety margin. |
| 1.67 | 5.0 | Less than 1 | Six Sigma target. The semiconductor industry sets this as the standard goal. Near-zero defect probability under normal operation. |
| 2.00 | 6.0 | ~0.002 | World-class. The process uses only 50% of the specification width. Defects are measured in parts per billion. |
A negative Cpk deserves special attention. It means the process mean has crossed a specification limit. You are not flirting with the edge of the tolerance. You are past it. Every part the process produces is suspect until the mean is brought back between the limits.
4. Why 1.33 and 1.67? The History Behind the Thresholds
Cpk = 1.33 did not become the industry standard by accident. The number corresponds to a 4-sigma safety margin: the distance from the process mean to the nearest specification limit is four standard deviations. At that distance, the defect rate is approximately 64 parts per million.
The threshold balances two competing forces. Push it higher and you spend more on tighter tolerances, better equipment, and more inspection than the defect reduction justifies. Let it slip lower and the defect rate climbs from 64 ppm at Cpk = 1.33 to roughly 2,700 ppm at Cpk = 1.0. A process that looked adequate can become a liability with a small shift in the mean.
Cpk = 1.67 is the stricter standard. In the semiconductor industry, where a single wafer contains hundreds of chips and a defect on any one of them scraps the entire unit, the Cpk goal is normally set at 1.67. A Cpk of 1.33 is still considered acceptable, but 1.67 is the target. At that level, the defect rate drops below 1 part per million.
5. What Cpk Tells You About Process Drift
Cpk is a leading indicator. When it drops, something is moving.
The direction of the drop tells you what to fix. If Cp is high and Cpk is falling, the process spread is fine. The mean is drifting toward one of the specification limits. Check calibration, check setpoints, check whether operators are targeting the true centre of the tolerance or favouring one edge.
If Cp and Cpk are falling together, the variation itself is increasing. Look at raw material consistency, tool wear, environmental conditions, and measurement system variation. This is a DMAIC Analyse-phase problem, not a quick adjustment.
If Cpk is below 1.0, you are already shipping defects. The process is not capable and output frequently falls outside specification limits. Stop and fix. A Cpk below 1.33 means the process is too close to its specification limits, making it vulnerable to shifts and variations that could lead to defects.
The relationship between Cp and Cpk is captured by Cpk = Cp(1−k), where k measures how far the process mean is from the midpoint of the specification range. A perfectly centred process has k = 0, so Cpk = Cp. An off-centre process has k > 0, so Cpk drops below Cp. The size of k is a direct measure of how much capability you are losing to poor centring.
6. Cpk vs Ppk: Short-Term Capability vs. Long-Term Performance
Cp and Cpk use short-term standard deviation. This is within-subgroup variation, calculated from consecutive parts or samples taken close together. Pp and Ppk use long-term standard deviation. This is the overall variation across the entire dataset, including shifts between batches, operators, and over weeks of production.
The distinction matters because it tells you whether your process is stable. When a process is in statistical control, Cpk and Ppk converge to nearly the same value. If Cpk is significantly higher than Ppk, special causes are present. Something is shifting the process between subgroups that does not show up within them. That gap is a signal to investigate.
Think of it this way: Cpk is the process on its best behaviour, during a single production run under consistent conditions. Ppk is the process as it actually performs over weeks and months, across shift changes, raw material batches, and tool wear cycles. If the two numbers are close, your process is stable. If they diverge, you have special causes to find and eliminate.
7. How to Read a Cpk Number in 30 Seconds
Here is a decision framework you can use without opening a textbook.
Cpk less than 1.0: Stop and fix. The process is not capable. Output is falling outside specification limits. You are shipping defects. The fix is not monitoring. It is an engineering change. Reduce variation, shift the mean, or widen the tolerance if the design allows it.
Cpk 1.0 to 1.33: Improve or monitor. The process is barely capable. A shift of half a standard deviation puts you into defect territory. If this is a critical characteristic, invest in improvement. If it is non-critical, increase monitoring frequency and set a control limit that triggers investigation before the Cpk drops below 1.0.
Cpk 1.33 to 1.67: Acceptable, monitor for drift. The process is highly capable, with 99.99% of output within specification limits. It has a safety margin to account for variability. Routine monitoring is sufficient. Track the trend. A Cpk that was 1.5 last quarter and 1.35 this quarter is a process that is walking toward a problem, even if it has not arrived yet.
Cpk above 1.67: Excellent, consider reducing inspection. The defect rate is below 1 part per million. You may be over-inspecting. Consider reducing sampling frequency and redirecting quality resources to processes with lower Cpk values.
The numbers are not a report card. They are a gauge. Read them that way.
If you are calculating process capability Cpk from raw data, the arithmetic is straightforward but easy to get wrong under time pressure. The formulas are simple: Cp = (USL − LSL) / (6σ) and Cpk = min[(USL − μ)/(3σ), (μ − LSL)/(3σ)]. The harder part is ensuring your data is normally distributed and your process is in control before the indices are valid. SimplicityHub's free Cp/Cpk calculator handles the arithmetic so you can focus on interpreting what the number is telling you about your process. For a deeper dive into the formulas and interpretation, see our process capability guide.