Your Cpk reads 1.5 on the monthly report and the customer signs off without another question. Three weeks later a complaint lands on the same characteristic, carrying the same number on the cover page. The Cpk was telling half the truth. The other number sitting right next to it never got read.
Process capability compares how much a characteristic varies against the tolerance the customer gave you. Cp tells you whether that variation fits inside the tolerance. Cpk tells you whether it fits and sits centered where it should.
The Cp that looks good and says nothing about where the process stands
Cp is calculated as Cp = (USL − LSL) / 6σ, where USL and LSL are the upper and lower specification limits and σ is the process standard deviation. It compares the width of the tolerance against six standard deviations, which is where almost all of the normal process variation falls. A Cp of 1.67 tells you the tolerance is generous compared to how much the process varies. It does not tell you whether the process sits in the middle of that tolerance or crowds one edge.
Cpk: the same variation, except it cares where it is standing
Cpk corrects for that. It is calculated as Cpk = the minimum of (USL − X̄) / 3σ and (X̄ − LSL) / 3σ, where X̄ is the actual process average. Instead of using the full tolerance, it measures the distance from the average to each limit separately and keeps the shorter one. When the process sits exactly centered, Cp and Cpk come out equal. The moment it drifts to one side, Cpk drops while Cp stays just as pretty.
The example I draw on the whiteboard
Take a part with a tolerance of 10.00 mm ± 0.10 mm, so LSL is 9.90 and USL is 10.10. If the process standard deviation is 0.02 mm, Cp comes out to (10.10 − 9.90) / (6 × 0.02) = 0.20 / 0.12 = 1.67. That reads excellent on any report.
Now assume the real process average sits at 10.06 instead of 10.00, drifted toward the upper limit. The distance to USL is 10.10 − 10.06 = 0.04, divided by 3σ (0.06) gives 0.67. The distance to LSL is 10.06 − 9.90 = 0.16, divided by 3σ gives 2.67. Cpk is the smaller of the two: 0.67. Same process, same gage, same day. One number passes, the other fails.
Why 1.33 became the number everyone asks for
As a general industry convention, a minimum Cpk of 1.33 is what most automotive and corporate customers ask for on standard characteristics, and 1.67 when the characteristic is safety or function critical. It is the margin a centered process keeps to absorb added variation without slipping outside tolerance. A Cpk of 1.00 sits right at the edge, and any small drift already produces a reject.
What to move when Cpk does not clear the bar
In the example above, the cheapest fix is almost never reducing variation. It is recentering the process: move the average back from 10.06 to 10.00 and Cpk climbs to 1.67 without touching a single machine parameter beyond the offset. Reducing σ costs money, time and sometimes new technology. Recentering is usually an adjustment.
Three Cpk mistakes I keep seeing on the floor
- Reporting Cp when the customer asked for Cpk. A high Cp next to a low Cpk hides exactly the problem the customer wants visibility on.
- Calculating from fewer than 30 pieces and treating the number as final. With that few parts, σ has not settled yet.
- Chasing σ before checking centering. The cheapest fix almost always sits in moving the average, not squeezing the variation.
Where this fits
Cpk is the data point behind the verification step in an 8D or the analysis box in an A3, and it is the first thing I check when a plant reports a recurring quality problem without knowing where to start. A misread Cpk produces the same failure as a poorly built fishbone diagram: something gets approved that is already failing on the floor.
Start here
Pick a characteristic your customer audits regularly and calculate both numbers, not one. If Cp comes out high and Cpk comes out low, you already know where the problem sits: in where you parked the process. That is the first data point I check in a problem solving workshop.
What is the difference between Cp and Cpk?
Cp compares the tolerance against the process variation without looking at where it is centered. Cpk makes the same comparison from the real average toward each limit, so it catches a process that has drifted off center.
What Cpk does a customer usually ask for?
As a general industry convention, 1.33 is the most common minimum for standard characteristics, and 1.67 for ones that are safety or function critical. Each customer can set a different number in its own specification.
Does a low Cpk always mean you need to reduce variation?
Not necessarily. Check centering first. Recentering the average is usually cheaper and faster than reducing the standard deviation, and it often fixes a low Cpk without touching σ at all.