What is a good Cpk value? Benchmarks, PPM and customer minimums.
Every capability report ends with the same question from the customer side of the table: is that Cpk good enough? The honest answer is that "good" is a contract term, not a statistics term — but the industry has settled on clear benchmarks. Here is what 1.0, 1.33, 1.67 and 2.0 actually mean in defects per million, who demands which number, and what to do when you miss.
The short answer
For an ongoing, statistically stable process, Cpk 1.33 is the accepted minimum across most of manufacturing. For new processes at initial approval — a PPAP initial study or a new-line qualification — customers typically demand 1.67. Safety-critical or special characteristics often carry 1.67 (sometimes 2.0) permanently. Anything below 1.0 means your process spread does not fit inside the tolerance even when perfectly centred on its own mean: you are shipping defects or sorting them out by inspection.
What each Cpk means in defective PPM
Cpk maps directly to an expected defect rate on the nearest specification limit, assuming a normal, stable process:
| Cpk | Sigma margin to nearest limit | Defective PPM (nearest side) | Verdict |
|---|---|---|---|
| 0.67 | 2σ | ~22,750 | Not capable — contain and fix |
| 1.00 | 3σ | ~1,350 | Marginal — any drift creates scrap |
| 1.33 | 4σ | ~32 | Standard minimum for ongoing production |
| 1.67 | 5σ | ~0.3 | Initial-study / safety-characteristic level |
| 2.00 | 6σ | ~0.001 | Six Sigma short-term capability |
The full conversion, including two-sided cases and the 1.5σ long-term shift convention, is worked through in the Cpk to PPM guide. To get your own number, feed your mean, standard deviation and limits into the free Cp/Cpk calculator — it returns Cp, Cpk and the expected PPM in one shot.
Where 1.33 and 1.67 come from
Cpk counts how many thirds of your process spread fit between the mean and the nearest limit. Cpk 1.0 means the limit sits exactly 3 standard deviations away. The classic minimums are just neat sigma margins: 1.33 is 4σ (3 × 1.33 ≈ 4), and 1.67 is 5σ.
The reason initial studies demand the higher number is uncertainty, not optimism. A 25-subgroup initial study is a snapshot: it has not yet seen tool wear across a full insert life, batch-to-batch material variation, monsoon-season humidity, or the night shift. Requiring 1.67 up front is a buffer so that when those sources of variation arrive, the process still clears 1.33. In my plant-head years, processes that scraped through qualification at 1.35 were the ones generating containment actions six months later; processes qualified at 1.7+ rarely came back to the table.
Typical customer minimums by industry
| Context | Typical requirement |
|---|---|
| Automotive PPAP initial study (AIAG) | Ppk ≥ 1.67 on special characteristics |
| Automotive ongoing production | Cpk ≥ 1.33 |
| Aerospace / defence key characteristics | Cpk ≥ 1.33, often 1.67 for flight-safety items |
| Medical devices (process validation) | Commonly Cpk/Ppk ≥ 1.33 with documented rationale |
| Six Sigma programme target | Cp = 2.0, Cpk ≥ 1.5 long-term |
| General engineering, non-critical dims | Cpk ≥ 1.0–1.33 by agreement |
Two cautions. First, these are conventions — your customer's SQM manual or drawing note is the contract, and Indian OEMs increasingly copy AIAG numbers verbatim into supplier quality manuals. Second, the initial-study requirement is usually written against Ppk, not Cpk — overall variation, not within-subgroup — and quoting the wrong index in a submission is a common rejection reason.
What to do when Cpk falls short
A low Cpk has exactly two mathematical causes, and the fix order matters:
- Centring (Cp high, Cpk low). If Cp is healthy but Cpk lags, the process is off-centre. Shifting the mean — a tool offset, a temperature setpoint — is cheap. Do this first.
- Spread (Cp itself low). If Cp is also poor, the variation is too large for the tolerance. That needs real engineering: machine capability, fixturing, material consistency, or a tolerance discussion with design.
- Measurement system. Before either, confirm the gauge. A gauge eating 30% of the tolerance inflates apparent spread and can single-handedly sink a capability study — run the numbers with a gauge R&R study.
- Containment meanwhile. Below 1.33 with customer parts moving, protect the customer: 100% inspection on that characteristic until capability recovers, with the derogation documented.
Can Cpk be too high?
Yes — economically. A Cpk of 3.0 on a non-critical dimension usually means one of three things: the tolerance is far looser than the process (fine, but stop measuring it so often), you are running a slower or costlier process than the drawing needs, or the data is suspect (truncated, over-filtered, or a gauge resolution artefact producing implausibly tiny sigma). Capability effort is a budget; spend it on the characteristics near the line, not the ones at 5σ comfort.
Common mistakes when judging Cpk
- Quoting Cpk from an unstable process. Capability indices assume statistical control; check the control chart first, or the number is fiction.
- Confusing Cpk with Ppk in submissions. Initial studies are judged on Ppk; ongoing SPC on Cpk. Label them correctly.
- Tiny samples. A Cpk computed from 10 parts has huge confidence intervals — a "1.4" could easily be a true 1.1. Use at least 100 readings / 25 subgroups for decisions.
- Ignoring normality. Runout, flatness and position are bounded at zero and not normal; transform or use appropriate methods before quoting PPM.
- Chasing the index instead of the process. Widening spec limits or trimming outliers improves the report, not the parts.
Capability data starts with clean inspection data
Every Cpk study begins with measured values traced to a specific drawing characteristic. If your CMM and shop-floor readings map to numbered balloons, capability studies assemble themselves; if they map to "that bore, I think", no statistics can save you. CadNexa's auto-ballooning tool converts a PDF drawing into a numbered characteristic sheet in minutes, so every reading lands against the right feature.
Related reading: the fundamentals in Cp vs Cpk explained, machine-level acceptance in Cmk calculation, and standard capability report formats on the templates page.