Somewhere along the way, statistics stopped being a tool on the molding floor and became a religion. The people who founded the discipline — Deming, Ishikawa, Juran, Taguchi — were clear that capability numbers were instruments for making better decisions, and just as clear that they were only useful when you understood what they were telling you. Then spreadsheets arrived, CpK became a single cell anyone could fill in, and a whole layer of “data dogs” grew up around numbers that very few people in the building actually understood.
The result is a familiar scene: a quality system that produces beautiful capability charts, a process that’s quietly drifting, and nobody connecting the two. CpK gets treated as a grade on a report card instead of what it is — a snapshot of how much room a process has between its natural variation and the edge of the spec.
This article doesn’t reproduce historical WJT Associates material. It applies the same shop-floor reasoning — use the number to make a decision, or don’t bother calculating it — to process capability in injection molding.
Cp and CpK, in plain language
Strip away the symbols and the two indices answer two different questions.
- Cp asks: is the process tight enough to fit inside the tolerance at all? It compares the spread of your variation to the width of the spec, and ignores where that spread is centered. Cp is the archery question: can you put a dozen arrows in a tight group?
- CpK asks the harder question: is that tight group actually centered on the target? It accounts for how close the process is to the nearer spec limit. You can shoot a beautiful 2-inch group and still miss the target by a foot — high Cp, low CpK. CpK is the number that catches that.
| Index | What it measures | What it misses | The archery analogy |
|---|---|---|---|
| Cp | Spread vs. tolerance width | Whether you’re centered on target | A tight group — anywhere on the wall |
| CpK | Spread and centering vs. nearer limit | Drift over time after the snapshot | A tight group on the bullseye |
A CpK of 1.0 means the process edge just touches the spec limit — every part should be in spec, with essentially no margin. The common 1.33 target builds in a safety band so that normal drift doesn’t immediately produce rejects. Anything below 1.0 means the process is producing out-of-spec parts as a matter of statistics, whether or not you’ve caught them yet.
The number you froze is lying to you
Here’s the failure that costs real money. A tool gets qualified, the CpK on critical dimensions comes in at a comfortable 1.5, the paperwork is signed, and that number gets treated as a permanent property of the job. Two years later the tooling has worn, the process has drifted, the parts are riding closer to a limit than anyone realizes — and the official CpK still says 1.5, because nobody re-measured it.
Process capability is not a constant. It’s a measurement of a process at a moment in time. Tooling wears, gates erode, cooling lines scale, check rings loosen — and capability degrades with all of it. A CpK that was real at qualification becomes fiction the day the process it described stops existing. The plants that get burned are the ones treating a two-year-old capability study as a current guarantee.
The discipline that fixes this isn’t more charts. It’s periodically re-measuring capability on the dimensions that matter and watching the trend. A CpK sliding from 1.5 toward 1.2 over six months is telling you the tool needs attention — long before it produces a single reject. That’s the signal worth paying for. A static number in a binder is not.
CpK as a profit tool, not a hurdle
There’s a flip side that quality-by-fear misses entirely. Capability can be used to improve profit, not just to clear a hurdle.
A process running at CpK 2.0 on a dimension is comfortably over-delivering — it has far more margin than the spec requires. That margin is sometimes real money left on the table:
- It may mean the process is being run conservatively (slow, cool, long cycle) to protect a tolerance that has room to spare. Some of that cycle time may be recoverable.
- It may mean a tolerance was specified far tighter than the part’s function actually needs, and a conversation with the customer could legitimately open it up.
- It may mean cavities are being scrapped or sorted against an internal limit stricter than the print, for no functional reason.
None of this means chasing the spec edge recklessly — CpK exists precisely so you don’t. It means treating a very high capability number as information: a signal that there may be cost to recover, the same way a low number is a signal of risk. Capability that only ever triggers “pass” and never triggers a decision is overhead.
| CpK reading | What it usually means | The decision it should trigger |
|---|---|---|
| Below 1.0 | Process is making out-of-spec parts statistically | Stop and fix — tooling, process, or spec is wrong |
| 1.0–1.33 | In spec but thin margin; drift will cause rejects | Tighten the process window or widen via tooling |
| 1.33–1.67 | Healthy, intended operating range | Maintain; watch the trend over time |
| Above ~2.0 | Over-delivering — lots of unused margin | Investigate recoverable cycle, cavity sorting, or spec |
Garbage in, confident garbage out
The most dangerous thing about a capability number is how authoritative it looks. A CpK reported to two decimals carries the air of precision — but it’s only ever as good as the measurements feeding it. A worn gauge, an inconsistent fixture, an operator measuring a flexible part with hand pressure that varies, a sample pulled all from one cavity on a multi-cavity tool — any of these produces a confident number that means nothing.
Before you trust a capability study, the unglamorous questions matter more than the math:
- Is the gauge capable and calibrated, and has the measurement system itself been checked for repeatability?
- Was the sample drawn across cavities and across time, or all at once from one cavity?
- Is the measured dimension actually the one that drives the part’s function?
- Are the data even in statistical control? Capability calculated on an out-of-control process is meaningless — you’re describing a process that doesn’t have a stable identity to describe.
A computer-generated answer is only as accurate as the inputs. A precise-looking CpK built on a sloppy measurement system is worse than no number at all, because people act on it.
FAQs
What’s the difference between Cp and CpK?
Cp measures whether your process variation is tight enough to fit inside the tolerance at all, ignoring where it’s centered — a tight group anywhere on the wall. CpK measures both the spread and whether that group is centered on target relative to the nearer spec limit — a tight group on the bullseye. You can have a high Cp and a low CpK if the process is tight but off-center, which is exactly the case CpK is designed to catch.
Is a higher CpK always better?
Higher is safer, but a very high CpK (say, above 2.0) can also be a sign you’re leaving money on the table — running conservatively, holding a tolerance tighter than the part’s function needs, or sorting cavities against an internal limit stricter than the print. CpK below 1.0 is a clear problem; a healthy operating range is usually around 1.33 to 1.67. Treat an unusually high number as information worth investigating, not just as a win.
Why does my CpK need to be re-measured?
Because process capability describes a process at a moment in time, not a permanent property of the job. Tooling wears, gates erode, cooling lines scale, and the process drifts — and capability degrades with all of it. A CpK that was accurate at qualification becomes fiction once the process it measured has changed. Re-measuring periodically and watching the trend gives you an early warning that a tool needs attention, often long before it produces a reject.
Can a CpK number be misleading even if it’s calculated correctly?
Absolutely. The arithmetic is only as good as the data feeding it. A worn or uncalibrated gauge, an inconsistent measurement fixture, a sample pulled all from one cavity, or data from a process that isn’t in statistical control will all produce a confident, precise-looking number that means nothing. Verify the measurement system and confirm the process is in control before you trust — or act on — any capability figure.