You've mapped the process. You've reduced waste. Cycle times are down, flow is smooth, and everyone feels good about the improvements.
Then a customer calls. The product is out of spec. Again.
Here's the uncomfortable truth: a process can be fast, efficient, and well-organized — and still incapable of meeting requirements. Speed and capability are different things. Process capability analysis tells you whether your process can consistently produce outputs within the boundaries that matter.
What Process Capability Means
Every process has natural variation. Parts come out slightly different sizes. Service times fluctuate. Chemical concentrations drift. This variation follows a distribution — often (but not always) a normal distribution with a center point and a spread.
Every output also has specification limits — the boundaries defined by the customer, the engineer, or the regulation. Upper Spec Limit (USL) and Lower Spec Limit (LSL). Anything inside those limits is acceptable. Anything outside is a defect.
Process capability compares your process variation to those specification limits. It answers a deceptively simple question: Does the process fit inside the spec?
The Key Metrics: Cp and Cpk
Cp (Process Capability Index)
Cp measures potential capability — how well the process could perform if it were perfectly centered.
Cp = (USL - LSL) / (6σ)
Where σ is the process standard deviation. The numerator is the spec width. The denominator is the process width (6 standard deviations covers 99.7% of output).
- Cp = 1.0: The process width exactly equals the spec width. Like parking a bus in a space exactly bus-length. Technically fits, but no room for error.
- Cp = 1.33: The spec is 33% wider than the process. Some breathing room. This is a common minimum target.
- Cp = 2.0: The spec is twice the process width. Six Sigma territory. Very capable.
- Cp < 1.0: The process is wider than the spec. You will produce defects, guaranteed, even if perfectly centered.
Cpk (Process Capability Index, adjusted for centering)
Cp assumes the process is centered between the spec limits. Cpk accounts for reality — processes drift off-center.
Cpk = min[(USL - μ) / (3σ), (μ - LSL) / (3σ)]
Cpk takes the worse side. If the process mean (μ) is shifted toward one spec limit, Cpk will be lower than Cp, reflecting the increased risk of defects on that side.
- Cpk = Cp: Process is perfectly centered. Ideal.
- Cpk < Cp: Process is off-center. The gap between Cp and Cpk tells you how much capability you're losing to centering problems.
- Cpk < 1.0: You're producing defects right now, regardless of what Cp says.
Why Both Numbers Matter
Imagine a process with Cp = 2.0 but Cpk = 0.8. The process has very little variation (good!) but it's centered way off-target (bad!). The fix is simple: re-center the process. Adjust the machine, recalibrate the instrument, retrain the operator. You don't need to reduce variation — just shift the mean.
Now imagine Cp = 0.9 and Cpk = 0.85. The process is nearly centered but the variation is too wide. Re-centering won't help much. You need to fundamentally reduce variation — better equipment, tighter controls, more consistent materials.
The combination of Cp and Cpk tells you what kind of improvement to pursue. That's what makes capability analysis actionable, not just diagnostic.
Beyond Manufacturing
Process capability isn't just for widget dimensions. Any process with measurable output and defined acceptable limits has capability:
- Service: Call handle time with a target of 3-7 minutes. Is your process capable of consistently hitting that window?
- Healthcare: Medication turnaround time with a 30-minute requirement. What percentage of orders exceed it?
- Finance: Loan processing within 48 hours. Is the process capable, or are you relying on expediting and heroics?
- Software: API response time under 200ms. Does the system consistently deliver, or do you have a fat tail of slow responses?
Anywhere you have a target and variation, you have a capability question.
Common Pitfalls
1. Assuming normality. Cp and Cpk assume normally distributed data. Many processes aren't normal — they're skewed, bimodal, or bounded. Always check with a histogram before calculating. Non-normal data needs different capability metrics.
2. Confusing Cp with Cpk. Reporting Cp alone is dangerously optimistic if the process isn't centered. Always report both.
3. Using short-term data. Capability calculated from one shift or one batch captures within-group variation but misses shift-to-shift and batch-to-batch drift. Use long-term data (Pp and Ppk) for the full picture.
4. Setting arbitrary targets. "We need Cpk > 1.33" is a reasonable general target, but the right target depends on the consequence of defects. Medical devices need higher capability than decorative trim.
5. Ignoring the voice of the process. If your process isn't capable, tightening the spec doesn't help — it just reclassifies more output as defective. Either improve the process or widen the spec.
Making Capability Actionable
Step 1: Collect data. Measure actual process output — at least 30 data points, ideally 100+, across multiple time periods.
Step 2: Check the distribution. Histogram, normality test. Understand what your data looks like.
Step 3: Calculate Cp and Cpk. Compare to your target (1.33 is a common minimum).
Step 4: Diagnose. If Cpk < target, is it a centering problem (Cp okay, Cpk low) or a variation problem (both low)?
Step 5: Improve. Center the process, reduce variation, or both. Then remeasure to confirm the improvement.
Process capability gives you the honest answer to a question most organizations avoid: Can we actually do what we're promising? If the answer is no, you know exactly where to focus. If the answer is yes, you have the data to prove it.
Either way, you stop guessing.