Your first pass yield is 98.2%. Your Cpk is 0.94. One says the process is fine. The other says it's incapable. Which metric is telling the truth?
Both are. They're answering different questions — and the confusion between and drives more bad decisions than most quality teams realize.
Your PPAP got rejected — not for bad parts, but for bad statistics. OEM auditors now scrutinize whether your Cpk method matches your data. Build a PPAP capability evidence chain that withstands the toughest audits.
First pass yield measures what already happened. Cpk predicts what will happen next. When they agree, your process is well-understood. When they disagree, something important is hiding in the gap.
First pass yield (FPY) is the percentage of units that pass inspection without rework, repair, or rejection on the first attempt. It's straightforward: 1,000 parts in, 982 pass, FPY = 98.2%.
FPY is a lagging indicator. It tells you what the process did — past tense. It's computed from completed production, after the variation has already happened and the defects have already been made.
Strengths of FPY:
Limitations:
Cpk measures how centered your process is within specification limits, scaled by the process spread. It's a leading indicator: it tells you how much room you have before defects start happening.
Cpk = 1.33 means the nearest specification limit is 4σ from your process center. Cpk = 1.0 means 3σ. Cpk = 0.67 means 2σ. The higher the Cpk, the larger the buffer between your process and the specification boundary.
Strengths of Cpk:
Limitations:
The interesting cases — and the decisions that go wrong — happen when FPY and Cpk tell different stories.
FPY = 99.1%. Cpk = 0.85. How?
The process is currently centered well within specifications — most parts pass. But the spread (σ) is large relative to the tolerance. Right now, the process mean happens to be in a good spot. A small shift — tool wear, material change, temperature drift — and defects will spike.
FPY says "everything's fine." Cpk says "you're one shift away from a problem."
The right response: Trust Cpk. Investigate the variation source. The yield number is temporarily masking an unstable situation. When the process drifts, yield will drop fast — and you won't have warning because you trusted the lagging indicator.
FPY = 94%. Cpk = 1.52. How?
Two explanations. First: the Cpk is wrong. If the data isn't normally distributed and you used standard Cpk, the number is unreliable. A skewed distribution can produce high Cpk while the tail generates defects that show up in yield.
Second: the defects aren't dimensional. FPY captures all failure modes — scratches, contamination, assembly fit issues. Cpk only covers the measured characteristic. The process might be dimensionally capable while failing on attributes Cpk doesn't measure.
The right response: Investigate which failure modes are driving the yield loss. If dimensional, recompute Cpk with distribution-free methods. If non-dimensional, Cpk is doing its job — the problem is elsewhere.
DPMO (Defects Per Million Opportunities) bridges yield and capability by converting both to the same scale.
Under normal-distribution assumptions:
| Cpk | Sigma Level | DPMO | Yield |
|---|---|---|---|
| 0.67 | 2σ | 45,500 | 95.45% |
| 1.00 | 3σ | 2,700 | 99.73% |
| 1.33 | 4σ | 63 | 99.9937% |
| 1.67 | 5σ | 0.57 | 99.99994% |
These conversions assume normality. For non-normal data, the DPMO calculated from standard Cpk doesn't match the actual defect rate — which is exactly the FPY/Cpk disagreement showing up in a different form.
When your observed DPMO (from actual defect counts) doesn't match the predicted DPMO (from Cpk), the distributional assumption is probably wrong. That gap is diagnostic: it tells you the normal model isn't capturing your process shape.
Many organizations convert Cpk to "sigma level" and report it as a performance metric. Cpk 1.33 = "4 sigma." Cpk 2.0 = "6 sigma." Clean, impressive, and frequently misleading.
The conversion assumes normality. If your Cpk was computed from non-normal data using the standard formula, the sigma level is doubly wrong — the Cpk is wrong, and the conversion from Cpk to sigma assumes a distribution that doesn't match.
If you must report sigma levels, compute them from actual DPMO (observed defects / opportunities), not from Cpk. The observed DPMO is distribution-independent — it counts what actually happened, regardless of what bell curve the process may or may not follow.
Neither FPY nor Cpk is universally "better." They answer different questions for different audiences:
Use FPY for:
Use Cpk for:
Use both together for:
FPY counts what happened. Cpk predicts what will happen. Neither is wrong — but both can mislead when their assumptions are violated.
FPY is honest by construction — it counts actual results. Cpk is honest only when the statistical method matches the data. For the 60–80% of industrial data that isn't normal, standard Cpk predicts a defect rate that doesn't match observed yield. Distribution-free Cpk closes that gap.
When FPY and Cpk finally agree, you know two things: your process is performing, and your statistical model is correct. That's the best position to be in.
Upload your data and see if your Cpk agrees with your yield — distribution-free capability that matches what you observe on the floor. Analyze your data free →
Minitab offers non-normal options. EntropyStat is distribution-free. Those aren’t the same thing. Offering a menu of distributions to choose from is distribution-flexible — not distribution-free. Here’s why that distinction determines whether your Cpk is correct.
With a small sample of 10 parts, traditional Cpk has a confidence interval 0.6 units wide — your 1.38 could be anywhere from 1.05 to 1.71. Entropy-based methods extract more from limited data without the normality assumption.
Process capability indices (Cpk and Ppk) quantify how well a manufacturing process can produce parts within specification limits. Cpk measures short-term capability using within-subgroup variation, while Ppk measures long-term performance using overall variation.
DPMO measures the number of defects expected per million opportunities for a defect to occur. It normalizes defect rates across products with different complexity levels, enabling fair comparison between a simple stamped bracket (few opportunities) and a complex PCB assembly (thousands of opportunities).
Yield analysis measures the proportion of products that pass all quality checks without rework or rejection. It includes first pass yield (FPY), rolled throughput yield (RTY), and final yield — each capturing different aspects of process quality across single or multiple manufacturing steps.
First pass yield measures the percentage of units that pass all quality checks on the first attempt without rework, repair, or rejection. It quantifies the true process quality by excluding the hidden factory of rework loops that inflate final yield numbers.