Lean Six Sigma · Beginner · 15 min read
Process Capability: Will Your Process Meet the Spec?
Lean Six SigmaYour customer sets the goalposts. Your process kicks the ball. Process capability measures — with two simple numbers, Cp and Cpk — whether your process can reliably land inside those goalposts.
Why this matters at work: capability indices are the standard language between suppliers and customers. A contract that says “maintain Cpk ≥ 1.33” is a precise, testable promise about defect rates — and knowing how to compute and read these numbers is a core skill for quality roles and a guaranteed topic on the ASQ CSSBB and CQE exams.
By the end, you should be able to
- Explain what process capability compares: the customer’s specification width against the process’s natural spread.
- Calculate Cp and Cpk from a mean, a standard deviation, and specification limits.
- Interpret the results against the common benchmarks (1.00, 1.33, 1.67) and spot an off-centre process.
- Run a capability analysis in Excel and Minitab and know which menu option to pick.
Before you start
This lesson builds directly on The Normal Distribution: Meet the Bell Curve. You should be comfortable with the mean μ, the standard deviation σ, and the idea that almost all values of a normal process fall within ±3σ of the mean. That “±3σ” span — 6σ wide in total — is the star of this lesson.
Two voices: the customer and the process
Every capability question is a conversation between two voices:
Voice of the customer — the specification
The customer (or the drawing, or the regulation) states what is acceptable: a lower specification limit (LSL) and an upper specification limit (USL). A coffee roaster promising “500 g ± 6 g” has set LSL = 494 g and USL = 506 g. The distance between them, USL − LSL, is the specification width — the width of the goalposts.
Voice of the process — the natural spread
The process doesn’t care what the drawing says. It produces values with its own mean μ and standard deviation σ. For a stable, roughly normal process, essentially all output (99.73%) lands within μ ± 3σ — a natural spread that is 6σ wide.
Capability is the comparison of the two widths. If the goalposts (spec width) are much wider than the ball’s scatter (6σ), nearly everything lands in-spec. If the scatter is as wide as the goalposts — or the aim is off-centre — defects follow.
Terminology
| Term | Meaning | Plain-language version |
|---|---|---|
| LSL / USL | Lower / upper specification limit | The goalposts set by the customer |
| Specification width | USL − LSL | How far apart the goalposts are |
| Natural process spread | 6σ (from μ − 3σ to μ + 3σ) | How widely the process scatters its output |
| Cp | Potential capability | Could the process fit, if perfectly centred? |
| Cpk | Actual capability | Does it fit, given where it’s actually aimed? |
| Pp / Ppk | Performance indices (overall σ) | Same idea, computed with long-term variation |
| Stable / in control | Only common-cause variation present | The process behaves the same way over time |
The formulas: Cp and Cpk
Cp is a simple ratio of the two widths: specification width on top, natural process spread underneath. Cp = 1 means the goalposts and the scatter are exactly the same width — the process barely fits, with zero room for error. Cp = 2 means the goalposts are twice as wide as the scatter. But Cp has a blind spot: it completely ignores where the process is centred. A process aimed way off-target can still have a beautiful Cp.
Cpk fixes the blind spot. It measures the distance from the process mean to the nearest spec limit, in units of 3σ, and reports the worse side. If the process is perfectly centred, Cpk = Cp. The further the mean drifts toward a limit, the lower Cpk falls — even while Cp stays put. That gap between the two numbers is itself a diagnosis:
Interactive: capability explorer
Move the sliders and watch both indices react. Try this experiment: leave the spread alone and slide the mean off-centre — Cp won’t budge, but Cpk will sink. Then re-centre the mean and widen σ instead — now both fall together.
Capable: Cpk meets the common 1.33 benchmark.
How to run a capability study, step by step
- Confirm the process is stable first. Run a control chart (see Control Charts Explained). Capability numbers from an unstable process are fiction — the spread you measure today won’t be the spread you get tomorrow.
- Check the data are roughly normal. A histogram or normal probability plot is enough at this level. Strongly skewed data need a transformation or a different method before the formulas apply.
- Collect enough data. Aim for at least 30 measurements — 50 to 125 is better — taken across a representative stretch of production, not one lucky hour.
- Compute μ and σ. Use the sample mean and sample standard deviation (or an estimate from a control chart’s subgroups — that distinction becomes Cpk vs Ppk later).
- Apply the formulas. Cp from the two widths; Cpk from the nearest-limit distance.
- Interpret and act. Compare against the agreed benchmark, then diagnose: centring problem, spread problem, or both.
Worked example: the coffee bag line
A packing line fills 500 g coffee bags with a spec of 500 ± 6 g (LSL = 494, USL = 506). Sixty bags from a stable run give a mean of μ = 501.2 g and a standard deviation of σ = 1.5 g.
- Specification width: 506 − 494 = 12 g.
- Natural spread: 6σ = 6 × 1.5 = 9 g.
- Cp = 12 / 9 = 1.33. Potential looks healthy: the goalposts are a third wider than the scatter.
- Upper side: (506 − 501.2) / (3 × 1.5) = 4.8 / 4.5 = 1.07.
- Lower side: (501.2 − 494) / 4.5 = 7.2 / 4.5 = 1.60.
- Cpk = min(1.07, 1.60) = 1.07.
The story the two numbers tell: the spread is fine (Cp = 1.33), but the line is overfilling by 1.2 g on average, pushing output toward the upper limit and dragging Cpk down to 1.07 — below the 1.33 benchmark. The fix isn’t a costly variation-reduction project; it’s re-targeting the filler back to 500 g. Centre it, and Cpk rises to meet Cp at 1.33 — and the overfill give-away stops too.
Reading the numbers
| Cpk | Roughly how many out of spec | Typical reading |
|---|---|---|
| < 1.00 | More than ~2,700 ppm | Not capable — the process cannot reliably hold the spec |
| 1.00 | ~2,700 ppm (centred) | Barely fits — any drift creates defects immediately |
| 1.33 | ~63 ppm | The classic minimum most customers require |
| 1.67 | ~0.6 ppm | Strong — typical target for safety-critical characteristics |
| 2.00 | ~0.002 ppm | “Six Sigma” capability with the mean on target |
One more distinction you’ll meet immediately in software: Cpk vs Ppk. Cpk uses the within-subgroup (short-term) σ estimated from a control chart; Ppk uses the overall (long-term) standard deviation of all the data. Ppk is usually a little lower, because long-term data picks up the small shifts between subgroups. Minitab prints both — now you know why they differ.
Common mistakes
Statistics implementation: Excel and Minitab
Software should confirm and speed up what you now understand — not replace the thinking.
Excel functions
Excel has no built-in Cp/Cpk function — you assemble them from these pieces:
| Function | Syntax | Purpose | When to use |
|---|---|---|---|
AVERAGE |
=AVERAGE(range) |
Process mean μ | Always — the first ingredient of every capability formula |
STDEV.S |
=STDEV.S(range) |
Sample standard deviation | Your default: measurement data is a sample of the process, not the whole population |
STDEV.P |
=STDEV.P(range) |
Population standard deviation | Only when the range truly is the entire population — rare in capability work |
MIN |
=MIN(a, b) |
Picks the worse side | The “min” in the Cpk formula: nearest-limit side wins |
NORM.DIST |
=NORM.DIST(x, mean, sd, TRUE) |
Probability of a value below x | Estimating the out-of-spec fraction below the LSL |
NORM.S.DIST |
=NORM.S.DIST(z, TRUE) |
Same, from a z-score | When you’ve already standardised (z = (limit − μ)/σ) |
COUNTIFS |
=COUNTIFS(range,"<"&LSL) + COUNTIFS(range,">"&USL) |
Actual out-of-spec count | Sanity-checking the model’s ppm prediction against the raw data |
Excel use cases
With data in A2:A61, and E1 = LSL, E2 = USL:
- Mean and spread:
=AVERAGE(A2:A61)inE4,=STDEV.S(A2:A61)inE5. - Cp:
=(E2-E1)/(6*E5)— the coffee-bag data returns 1.33. - Cpk:
=MIN((E2-E4)/(3*E5), (E4-E1)/(3*E5))— returns 1.07, flagging the off-centre fill. - Predicted out-of-spec:
=NORM.DIST(E1,E4,E5,TRUE) + (1-NORM.DIST(E2,E4,E5,TRUE)), formatted as a percentage (multiply by 1,000,000 for ppm).
Minitab navigation
| Menu path | What it gives you | When to choose it |
|---|---|---|
| Stat › Quality Tools › Capability Analysis › Normal | The classic single report: histogram with spec limits, Cp/Cpk (within) and Pp/Ppk (overall), plus expected ppm | Your standard choice once stability is already confirmed — the report customers expect |
| Stat › Quality Tools › Capability Sixpack › Normal | Control charts, normality plot, and capability in one six-panel view | Best first run: it checks the stability and normality assumptions and the capability together |
| Stat › Quality Tools › Capability Analysis › Nonnormal | Capability using a fitted distribution (e.g. Weibull, lognormal) | When the normality check fails and a transformation isn’t appropriate |
| Graph › Probability Plot | Quick normality check on its own | Fast assumption screening before any capability run |
In the dialog, enter your data column, the subgroup size (enter 1 for individual measurements), and the
LSL/USL. Minitab estimates the within-subgroup σ for Cpk and the overall σ for Ppk —
which is exactly why the two sets of indices differ on the report.
Exam tips (ASQ CSSBB / CQE)
- Memorise both formulas cold. Exam questions hand you μ, σ, LSL, USL and expect Cp and Cpk in under a minute. Practice until the coffee-bag example takes you 30 seconds.
- Know the signature patterns: Cp = Cpk ⇒ perfectly centred; Cp > Cpk ⇒ off-centre; Cpk can never exceed Cp. A negative Cpk means the mean is outside a spec limit.
- One-sided specs: with only a USL, Cpk = (USL − μ)/3σ and Cp is undefined — a favourite exam trap.
- Short-term vs long-term: Cp/Cpk pair with within-subgroup (short-term) σ; Pp/Ppk with overall (long-term) σ. Questions about “which index uses which sigma” appear on both exams.
- Benchmarks: 1.00 ≈ 2,700 ppm, 1.33 ≈ 63 ppm, 1.67 ≈ 0.6 ppm (centred, two-sided). Knowing 1.33 ⇔ “4σ to the nearest limit” lets you re-derive the rest.
- Prerequisites are testable: “What must be true before a capability study?” — stability (in control) and approximate normality. Stability always comes first.
Try it yourself
A machining line cuts shafts to a diameter spec of 10.0 ± 0.5 mm (LSL = 9.5, USL = 10.5). A stable run gives μ = 10.1 mm and σ = 0.12 mm. Compute both indices to two decimal places.
- Spec width = 10.5 − 9.5 = 1.0 mm; natural spread = 6 × 0.12 = 0.72 mm.
- Cp = 1.0 / 0.72 = 1.39.
- Upper side: (10.5 − 10.1) / 0.36 = 1.11; lower side: (10.1 − 9.5) / 0.36 = 1.67.
- Cpk = min(1.11, 1.67) = 1.11 — the spread is fine, the aim is 0.1 mm high.
Check your understanding
Five questions. Pick an answer for each, then press Check my answers.
Final summary
- Capability compares the voice of the customer (spec width) with the voice of the process (6σ spread).
- Cp = (USL − LSL) / 6σ measures potential; it ignores centring.
- Cpk = min[(USL − μ)/3σ, (μ − LSL)/3σ] measures reality; it reports the worse side.
- A gap between them is a diagnosis: re-aim if only Cpk is low, reduce variation if both are.
- Benchmarks: 1.00 barely fits, 1.33 the classic requirement, 1.67 safety-critical territory.
- Stability first, capability second — and check normality before trusting the ppm predictions.
References & further reading
- NIST/SEMATECH e-Handbook of Statistical Methods — Process Capability: itl.nist.gov/div898/handbook
- ASQ Quality Resources — Process Capability (Cp, Cpk): asq.org/quality-resources
- Montgomery, D. C., Introduction to Statistical Quality Control — the standard reference chapter on capability indices.