A capability index is one of the most trusted numbers in quality. A Cpk of 1.67 on a report reads as proof that a process is in control and comfortably meeting its requirement. Customers ask for it, PPAP packages depend on it, and a good number tends to end the conversation. That trust is exactly the problem, because a capability index is only meaningful under conditions that are quietly violated all the time. Break any one of them and the impressive number on the page is telling you almost nothing, or worse, telling you something false.
capability assumes stability
The first assumption underneath any capability number is that the process was stable when you measured it. Capability describes the expected performance of a predictable process, and a process that is drifting, shifting, or occasionally jumping is not predictable. Calculate a capability index on an unstable process and the math will still produce a number, a confident-looking figure that describes nothing real, because there is no single stable behavior for it to describe. A great Cpk on a process that is not in statistical control is not good news. It is a meaningless calculation dressed up as a result, and it hides the instability instead of revealing it.
capability assumes the measurement is trustworthy
The second assumption is that the measurement system telling you the values can actually be trusted. Every capability number is built on measured data, and if the gauge and method carry significant variation of their own, that measurement variation is baked into the result. A capability index can look poor purely because the measurement is noisy, sending teams chasing a process problem that does not exist. It can also look artificially fine when the measurement is masking real variation. Without knowing the measurement system is sound, the capability number is a mix of process reality and measurement error in unknown proportions, and you cannot tell how much of each you are looking at.
capability assumes the sample represents production
The third assumption is that the parts you studied actually represent the real production process. A capability study run on a short, carefully supervised run, with the best operator, the fresh tooling, and the ideal material, will produce a number that has little to do with what happens across shifts, operators, batches, and months of real production. The study captured a best case and presented it as the norm. This is one of the most common ways capability numbers mislead, not through bad math, but through a sample that quietly excludes all the sources of variation that make real production harder than the study.
why the false confidence is expensive
The danger of a capability number is precisely that it looks authoritative. A specific figure carried to two decimals invites people to stop asking questions, to treat the process as understood and move on. When that number rests on a process that was not stable, a measurement that was not verified, or a sample that did not represent production, the false confidence is worse than no number at all, because it actively discourages the scrutiny the situation actually needs. Teams relax on a process that is not really capable, and are then surprised when it produces defects the number said were nearly impossible.
earn the number before you trust it
None of this makes capability indices useless. Under the right conditions they are genuinely powerful. It makes them conditional, and the conditions are the whole point. Before trusting a capability number, confirm the process was stable, confirm the measurement system was sound, and confirm the study captured the real range of production rather than a best case. A capability index that has earned those three conditions is worth a great deal. One that has not is decoration, and the more impressive it looks, the more effectively it hides whatever it is failing to show.
When someone hands you a strong capability number, do you know whether the process was stable, the measurement was trustworthy, and the sample was real? Those three questions decide whether the number means anything at all.
I write about quality, manufacturing, and the lessons the floor teaches. If this resonated, follow along on LinkedIn and tell me what it brought up for you.
Connect on LinkedIn →