Cpk is defined by the closer specification limit, so the direct conversion describes that nearest tail. Total observed or two-sided PPM requires both specification limits and the process mean and sigma—or an explicit centered-process assumption.

Capability Conversion

PPM to Cpk Conversion

4 min read Last updated July 2026

PPM and Cpk are two views of the same thing under a normal distribution. The conversion is widely used in PPAP, SPC, and Six Sigma reporting.

Open the Process Capability Calculator Quality Statistics Knowledge Center

The engineering question this page answers

Given an observed or predicted PPM, what Cpk should I report — and under which assumptions is that number honest?

Decision logic

Classify the PPM: observed (from production) or predicted (from a normal model) ↓ Decide side: nearest-tail (one-sided) or total two-sided ↓ Only apply Z = NORMSINV(1 − PPM/1e6) after confirming normality on the same dataset ↓ Cpk = Z / 3 for nearest-tail; for two-sided centered use Z from PPM/2 ↓ State whether a 1.5σ shift was applied — never leave it implicit

Capability study readiness

Typical PPM and sigma equivalent

Typical PPM: Ref: PPM 1350 → Cpk 1.00 · PPM 32 → 1.33 · PPM 0.29 → 1.67 · PPM 0.0009 → 2.00 (nearest tail, no shift).

Formula

Nearest-tail: Cpk = NORMSINV(1 − PPM/1e6) / 3 Two-sided (centered): Cpk = NORMSINV(1 − PPM/2e6) / 3 With 1.5σ shift: Cpk_LT = (Z_ST − 1.5) / 3

Examples

Example. Assembly line reports 233 predicted PPM on the nearest tail, normality confirmed, process stable.

p = 233 / 1e6 = 0.000233
Z = NORMSINV(0.999767) ≈ 3.50
Cpk ≈ 3.50 / 3 = 1.17
Engineering conclusion: below AIAG 1.33 minimum; centering or variation reduction required before PPAP.

Conversion conditions

Distribution modelNormal distribution, justified from stable study data.

Specification modelcontext-dependent

Tail conventioncontext-dependent

Centering assumptionCentering and tail allocation must be stated; total PPM alone is not generally sufficient.

Sigma basisuser-specified

Sigma-shift conventionnone

Output typepredicted

Required inputs

PPM value
whether PPM is one-sided, nearest-tail, or centered two-sided
distribution and stability basis

Invertibility. Invertible only after the tail convention and centering model are declared.

Limitations

Asymmetric tail allocation cannot be recovered from total PPM alone.
The result is a normal-model prediction, not observed defects.

Worked example and engineering limits

Worked example. For 66.1 centered two-sided predicted PPM under a normal, centered, no-shift model, use 33.05 PPM per tail: Cpk is approximately 1.33.

Failure case. The same total PPM split unequally between tails does not identify one unique Cpk.

What cannot be inferred. Observed defect rate, stability, centering, or both tail distances.

Engineering next step. Inspect the control chart and retain both tail distances before using the conversion for a customer decision.

Engineering procedure

Confirm process stability (control chart in control) before any conversion.
Test normality (Anderson–Darling or probability plot). If it fails, stop — the conversion is invalid.
Compute the tail probability p = PPM / 1,000,000 for nearest-tail, or p = PPM / 2,000,000 for a symmetric two-sided assumption.
Z = NORMSINV(1 − p). Report Cpk = Z / 3.
Label the result: "predicted, nearest tail, no shift" or equivalent — never a bare number.

Common mistakes

Converting observed short-run PPM as if it were a stable long-run rate.
Mixing two-sided PPM into a one-sided Cpk formula — inflates Cpk by roughly Z(p) − Z(p/2).
Applying the 1.5σ shift silently, then comparing to non-shifted customer targets.
Treating attribute defect rates (pass/fail) as if they belonged to a continuous normal model.
Ignoring measurement error — MSA contamination inflates predicted PPM.

Engineering insight

PPM-to-Cpk is a model, not a measurement. The conversion inherits every assumption of the normal distribution and every weakness of the sample. If any assumption fails, the number produced is precise but not accurate.

When NOT to use this metric

Non-normal data (skewed, bounded, bimodal) — use Johnson/Box-Cox or a non-normal capability method.
Attribute data — use p-chart based capability, not a Z-transform.
Autocorrelated data — the σ estimate is biased; time-series modelling required first.
Sample size < 30 subgroups or < 100 individuals — the conversion has wide confidence bounds.
Unstable processes — capability is undefined; stabilize first.

Relationship to other capability metrics

PPM ↔ Sigma Level: Z is the linking quantity; Cpk is Z / 3.
Observed vs predicted PPM: Divergence usually flags non-normality or instability, not process improvement.
Cp vs Cpk from PPM: PPM converts to Cpk (nearest tail), never directly to Cp — Cp requires the spec width and σ.

Recommended actions

Use the free Process Capability Calculator to compute Cp, Cpk, Pp, Ppk, sigma level, and predicted PPM from your measurement data — no sign-up required.

Continue the investigation

obtaining tail and centering evidence — Conversion assumptions require study evidence.

Read the inverse in Cpk → PPM, the Z bridge in Sigma → Cpk, and the master reference Process Capability Formula. For attribute or non-normal data, review Process Capability Analysis.

Verification checklist

Process stable on control chart
Normality tested and passed on the same dataset
Tail side (nearest / two-sided) stated explicitly
1.5σ shift disclosed or explicitly absent
MSA acceptable — measurement error separated from process variation

Assumptions and applicability

Process condition: statistical stability is required.
Distribution assumption: approximately normal data.
PPM convention: nearest specification-limit tail.
Sigma-shift convention: no sigma shift applied.
Cpk alone identifies distance to the nearest specification limit; it does not determine total two-sided defects unless centering or both tail distances are known.
Predicted PPM is a model estimate, not a replacement for observed defect data.

Sources and engineering references

External engineering references used for this page. Qhubio applies these references to the practical guidance above.

External engineering referenceNIST/SEMATECH
What is Process Capability?
Definitions and interpretation of capability indices.
External engineering referenceNIST/SEMATECH
Normal Distribution
Normal-distribution assumptions used in PPM conversions.