Cpk Below 1: Risk, PPM, and Recovery

Cpk is below 1 on a stable, measured process. What is the shortest path back to a defensible index?

Cpk 1.00 predicts about 1,350 PPM beyond the nearest limit; 2,700 PPM applies only to a perfectly centered two-sided process.

• Typical PPM: Cpk 0.50: ≈ 66,807 PPM at the nearest limit; ≈ 133,614 PPM only for a perfectly centered, two-sided process; Cpk 0.80: ≈ 8,198 PPM at the nearest limit; ≈ 16,395 PPM only for a perfectly centered, two-sided process; Cpk 1.00: ≈ 1,350 PPM at the nearest limit; ≈ 2,700 PPM only for a perfectly centered, two-sided process
• Sigma equivalent: Roughly 1.5σ – 3σ depending on Cpk

1. Apply containment per the Control Plan reaction plan.
2. Split the dataset by stream and recompute — pooled Cpk hides lane-to-lane bias.
3. Compute k = |μ − target| / (tolerance/2) to quantify centering loss.
4. Estimate variation-only Cpk = Cp; the delta from actual Cpk is the centering headroom.
5. Decide corrective path: re-center, reduce variation, or redesign.
6. Run a small verification study (n ≥ 30, stable) before returning to normal production.
7. Update reaction limits so a repeat drift is detected before it produces defects.

• "Improving" Cpk by removing tail measurements without a cause code.
• Re-tuning the machine mid-study to hit a target Cpk; the study becomes invalid.
• Adopting a tighter internal spec than the drawing to force a passing PPM.

• Cpk 0.8 to 1.0 is usually a centering problem, not a variation problem — the fastest fix is an offset, not a DoE.
• Cpk under 0.5 almost always requires a design or technology change; SPC tuning rarely recovers more than 0.3 in Cpk.
• Improvements that look large in a short window (a shift, a lot) often revert; re-verify after one full tool life.

• Do not use Cpk < 1 to compare processes — the tail model is unreliable this far from the ideal.
• Do not report Cpk when the data comes from an out-of-control chart.

• At Cpk 0.8, PPM(nearest) ≈ 8,200; at Cpk 1.0, PPM(nearest) ≈ 1,350.
• The gap to the customer target (usually 1.33) sets the required improvement in σ or in mean.

1. Containment until Cpk is restored.
2. Separate the centering problem from the variation problem.
3. Run a designed experiment if root cause is unclear.
4. 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.

• Never celebrate an improvement below 1.33 — the process is still producing measurable defects.
• A Cpk between 0.9 and 1.1 on repeated studies is a signal of chronic drift, not random noise.

• ✓Stability and MSA re-verified before recomputation
• ✓Streams separated before pooling
• ✓Centering vs variation loss quantified
• ✓Corrective action linked to the dominant loss, not to both blindly
• ✓Verification study run after the change

• 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.