Cpk (process capability index) is a statistical measure of how reliably a process produces within its tolerance limits, taking into account how far the mean has drifted from the centre. Cp, by contrast, compares only the process spread with the tolerance width. A Cpk of 1.33 or more is considered capable in most industries; below 1.00, the process is producing defects.
Key takeaways
- Cp compares the tolerance width with the process's 6σ spread; Cpk measures the distance from the mean to the nearer tolerance limit in units of 3σ.
- A Cpk of 1.33 is the common minimum; in automotive, customers often require 1.67 for new processes and critical characteristics.
- Cpk is only meaningful for a process shown to be stable on a control chart: achieve stability first, then measure capability.
- For a reliable Cpk, common practice is at least 25 subgroups of 4–5 parts, roughly 100–125 readings.
- The gap between Cp and Cpk shows how much capability you can gain simply by re-centring the process.
What is SPC, what is Cpk and how do they connect?
Statistical process control (SPC) is a set of methods that uses measurements taken during production to check whether a process is stable, and to flag problems before defective parts appear. It asks two questions:
- Is the process stable? In other words, is there only natural (common-cause) variation, or has a special cause such as tool wear, a new material lot or a settings drift come into play? The tool for this is the control chart.
- Does the stable process meet the tolerance? The tools for this are capability indices such as Cp and Cpk.
The order matters. A Cpk calculated for an unstable process predicts nothing. In SPC, stability comes first, via the control chart; capability is measured afterwards.
Within the quality and traceability stage of our Industry 4.0 roadmap for SMEs, SPC is the first and most tangible use of measurement data collected digitally from the floor. As long as readings are gathered by hand and late, the control chart lags behind too.
What does unmanaged process variation cost?
Process variation builds up quietly, and the invoice usually arrives in the field, in the customer's hands.
In a survey of more than 750 quality leaders in the US, UK and Germany carried out for ETQ, 73% had experienced at least one product recall in the last five years, and 48% said they now see more recalls than five years ago. (ETQ Pulse of Quality in Manufacturing 2024, via PR Newswire)
Warranty costs point the same way. According to Warranty Week's 23rd Annual Product Warranty Report, US-based manufacturers paid $30.37 billion in warranty claims in 2025, up 4% on 2024, with an average claims rate of 1.30% of product sales.
What stands out is that the data often already exists. In the same ETQ survey, 80% of respondents said most or all functions can access relevant quality data. The problem is not a lack of measurement but that measurement never turns into a control chart and a decision. When readings are written on forms and filed away, a trend is only spotted when a customer complains.
How do you calculate Cp and Cpk? Formulas and a worked example
The definitions used in Cp and Cpk calculations:
- USL / LSL: upper and lower specification (tolerance) limits.
- μ: process mean.
- σ: process standard deviation. Short-term (within-subgroup) σ is usually estimated as R̄ / d₂; for subgroups of five, d₂ = 2.326.
Formulas:
- Cp = (USL − LSL) / 6σ: how many times does the process spread fit into the tolerance?
- Cpk = min[(USL − μ) / 3σ, (μ − LSL) / 3σ]: how many "3σ" steps is the mean from the nearer limit?
Example: a shaft diameter is specified at 20.00 ± 0.05 mm, so USL = 20.05 and LSL = 19.95. Measurements show a mean of 20.01 mm and a standard deviation of 0.01 mm.
- Cp = (20.05 − 19.95) / (6 × 0.01) = 0.10 / 0.06 = 1.67
- Upper side: (20.05 − 20.01) / 0.03 = 1.33; lower side: (20.01 − 19.95) / 0.03 = 2.00
- Cpk = min(1.33, 2.00) = 1.33
Interpretation: the spread is comfortably narrow for the tolerance (Cp 1.67), but the mean has drifted 0.01 mm towards the upper limit. Re-centring the machine lifts Cpk to 1.67 without any investment. The gap between Cp and Cpk tells you what centring alone can gain.
In injection moulding, drifts like this often become visible once they are tracked alongside mould and cycle data; see injection moulding production monitoring.
Approximate equivalents of Cpk values, assuming a normal distribution:
| Cpk | Distance from mean to nearer limit | Approximate defect rate on that side (ppm: parts per million) | Interpretation |
|---|---|---|---|
| 1.00 | 3σ | ~1,350 ppm | Borderline; defects are inevitable |
| 1.33 | 4σ | ~32 ppm | Common minimum in many industries |
| 1.67 | 5σ | ~0.3 ppm | Common automotive target for new processes and critical characteristics |
| 2.00 | 6σ | ~0.001 ppm | Very high capability |
The values in the table are calculated theoretical rates; if the distribution is not normal, or the process is unstable, the real defect rate will differ. The table still helps when briefing management: raising Cpk from 1.00 to 1.33 cuts the theoretical defect rate on the near side roughly fortyfold. That makes Cpk improvement one of the few quality measures you can tie directly to scrap and rework cost.
What is the difference between Cpk and Ppk?
Cpk uses short-term (within-subgroup) σ and shows the process's potential. Ppk uses the overall standard deviation of all the data and reflects the performance the customer actually receives. If Ppk is well below Cpk, something is shifting between subgroups: shifts, lots or tool changes.
How do you implement SPC on the shop floor? Six steps
- Choose the critical characteristics. Do not chart everything. Start with special characteristics flagged on the customer drawing and dimensions that have drawn complaints.
- Validate the measurement system. A Cpk calculated without knowing the gauge's repeatability and the difference between operators (MSA, measurement system analysis / Gage R&R) is misleading.
- Define subgroups and sampling. Common practice is at least 25 subgroups of 4–5 parts (100–125 readings). Each subgroup should be made in a short time under the same conditions.
- Pick the right chart. X̄-R or X̄-S for variable data, I-MR for individual readings; p or np for proportion defective, c or u for defect counts.
- Achieve stability, then calculate Cpk. A point outside the control limits, eight consecutive points on one side of the centre line (Western Electric rules; nine under the Nelson rules) or a steady upward or downward trend signal a special cause. Find it, remove it, then measure capability.
- Write the reaction plan. When a chart signals, what does the operator do: stop the machine, adjust it or call quality? For recurring problems, get to the root cause with the 8D problem-solving method.
How to read a control chart
A control chart plots measurement averages over time with three lines: the centre line (the mean) and upper and lower control limits, usually ±3σ from the centre. In a stable process, points scatter randomly within that band with no pattern.
For operators the practical rule is simple: while the chart is quiet, leave the machine alone. Adjusting after every small deviation (over-adjustment) increases variation rather than reducing it. Intervene only when the chart signals a special cause, and note the action on the chart, so the answer to "what did we do last time?" is ready when the next drift appears.
In SPC, one distinction matters above all: control limits are calculated from the process's own data and are the voice of the process; specification limits come from the customer. Mixing the two on the same chart is the most common mistake in SPC.
Common mistakes in SPC and Cpk calculation
- Manual, delayed data. Readings written on a form at the end of the shift show a drift hours after it happened, not when it happened; the layer that records each reading against its work order and lot in real time is the MES, or manufacturing execution system (MES vs ERP).
- Deciding on too little data. A Cpk based on 10–15 readings carries very wide uncertainty.
- Ignoring non-normal distributions. One-sided characteristics such as surface roughness, runout or flatness are often skewed and need a transformation or a suitable distribution model.
- Separating process data from lots. If you do not know which lot, machine and tool a reading belongs to, root cause analysis stalls; our production traceability guide explains how to build that link.
For defects that need visual inspection, computer vision quality inspection can complement SPC, and linking machine-driven variation to condition data such as vibration analysis speeds up the search for root causes.
In automotive supply, SPC pays off when measurements are captured at the station and linked to the batch number; see digital transformation for automotive suppliers.
How does Digital Bridge set up SPC and Cpk tracking?
SPC earns its keep when a reading becomes a chart point instantly and the chart becomes a decision. At Digital Bridge, we build that chain with you:
- We digitise quality data. In quality management system projects we bring inspection plans, control charts and Cp/Cpk/Ppk calculations into one system, and the reaction plan appears on screen when a chart signals.
- We capture measurements at source. Readings from digital callipers, micrometers or machine sensors are recorded without manual entry, using industrial data terminals at the line.
- We put each reading in production context. Through MES production management every measurement is linked to its work order, lot, machine and operator. Our MES integrates with Logo, SAP, Mikro and bespoke ERP systems.
- We make it visible to management. Cpk trends by characteristic sit on a BI dashboard alongside OEE and scrap data. For help choosing the right measures, see our BI dashboard KPIs article.
In sectors with demanding customer audits, such as automotive suppliers, pulling capability reports straight from the system also saves a great deal of preparation time. Control plans and inspection instructions can be kept in SmartFiles.com.tr, Digital Bridge's business file management service in the Smart360 family, where each upload under the same name to the same folder becomes a new version and earlier versions can be restored. You can buy SmartFiles through Smart360.
To bring SPC data together with CAPA (corrective and preventive action), document control and calibration, see our quality management system software guide; for linking measurements to batch numbers in automotive, read IATF 16949 traceability requirements.
Where should an SPC project start?
Pick the one dimension that generates the most complaints or scrap and gather its readings from the last month. Is the data incomplete, handwritten or disconnected from lots? That single characteristic shows where an SPC project should start. Then contact us with what you find, and we will review your measurement points together in a free site assessment.
Our other quality, maintenance and production management guides are gathered in our complete Industry 4.0 guide.