Lexicon · The critique

What is Six Sigma?

A quality method, developed at Motorola in the 1980s, that cuts the variation in a process that already exists, project by project, towards 3.4 defects per million. It earns its keep at volume. It optimises the past; it does not create the next process, and a belt certifies the training, not the capability.

Read it as

Sigma level, Cpk and defects per million

Everything a process makes varies, if it is measured closely enough. Sigma, the standard deviation, is the size of that variation. A process is called six sigma when six standard deviations fit between its average and the nearer specification limit. Divide that distance by three and you have Cpk, the capability index a supplier quotes to a customer: four sigma is a Cpk of 1.33, six sigma a Cpk of 2.

Six Sigma the method was developed at Motorola in the mid-1980s, with the engineer Bill Smith widely credited for the core idea. Motorola was one of the first three winners of the Malcolm Baldrige National Quality Award, in 1988, and General Electric made the method famous after 1995. It runs improvement as projects in five phases, Define, Measure, Analyse, Improve and Control, known as DMAIC, led by staff trained to grades named after martial-arts belts. Six Sigma is a registered trade mark of Motorola.

No two parts come out of a machine exactly the same. Sigma measures how much they differ. If the average part sits six of those sigma steps away from the nearest limit the customer will accept, the process is called six sigma, and almost nothing it makes is rejected. Cpk says the same thing as one number: six sigma is a Cpk of 2.

Six Sigma is also the name of a method for getting there. It started at Motorola in the 1980s, General Electric made it famous, and the people trained in it are graded by coloured belts.

The boardThe bell curve, one ball at a time
0balls dropped, n
0inside the lines
0outside, in red
0outside per million, so far

Each ball is a part. At every peg it bounces left or right, the way a part picks up small errors on its way through a process. The two lines are the limits the customer will accept.

Drop a ball into a board of pegs. At every peg it bounces left or right, and it lands somewhere along the bottom. Drop a thousand and they pile up into a hill: most near the middle, a few out at the edges. That hill is the bell curve, and every machine in every factory draws one, part by part.

Now draw two lines either side of the middle: the limits the customer will accept. Balls that land outside them are scrap. Six sigma means the hill is so narrow that six of its widths fit between the middle and the nearer line. Then almost nothing lands outside.

Exhibit 1Sigma, Cpk and defects per millionHow tight the process is, and how many parts it rejectsHow narrow the hill is, and how many balls go over the line
6sigma to the nearer limitspreads between the middle and the nearer limithill widths to the nearer line
2.00Cpkcapability score (2 is excellent)capability score (2 is excellent)
3.4out of spec, per millionrejects per millionballs over the line, per million
1 in 294,000parts out of specparts rejectedballs over the line

Drag the slider, or switch the drift off and on. Calculated from the normal distribution.Drag the slider to tighten the process, and switch the drift to see what happens when it wanders off centre. The numbers are calculated, not illustrative.Drag the slider to narrow the hill, and switch on the wander to see the hill slide towards a line. The numbers are calculated, not illustrative.

The curve narrows as the sigma level rises from one to six. At six sigma held centred, about 0.002 parts per million fall outside the limits. Drifted by 1.5 standard deviations, the same process loses 3.4 per million.

Sigma level, Cpk and defects per million, centred and after a 1.5 sigma drift
Sigma levelCpkOut of spec per million, centredPer million after a 1.5 sigma drift
2 sigma0.6745,500308,538
3 sigma1.002,70066,807
4 sigma1.33636,210
4.5 sigma1.506.81,350
5 sigma1.670.57233
6 sigma2.000.0023.4

Centred counts both limits. The drift column counts the nearer limit only, as the standard tables do. Cpk is the distance from the average to the nearer limit, in standard deviations, divided by three.

The headline number is not six sigma. The famous 3.4 defects per million assumes the process average drifts by 1.5 standard deviations over time, so a six sigma process is scored as if it were four and a half. Held centred, six sigma loses about 0.002 parts per million, two in a billion. The same drift turns a Cpk of 1.33, the figure many customers write into supply contracts, from 63 parts per million into 6,210. The 1.5 is a convention drawn from Motorola’s experience, not a law of statistics, and it is still argued over.

The famous figure, 3.4 faulty parts per million, quietly assumes the process wanders off centre over time by a fixed amount, 1.5 sigma. Without that allowance a true six sigma process would lose about two parts in a billion. The allowance is a rule of thumb from Motorola, not a law of nature.

Machines wander. A tool wears, a batch of material changes, a new shift sets up a little differently, and the whole hill slides sideways. Press Add the drift on the board and watch the hill move towards one line and balls start falling over it. Six Sigma builds that wander in, which is why its famous 3.4 faulty parts per million is really the score of a process that has drifted by one and a half widths. Then press Tighten the process: a narrow hill can drift and still keep its balls inside.

Where Six Sigma earns its keep

Six Sigma is a statement about millions. Showing 3.4 defects per million takes millions of opportunities, a design that has stopped changing and a process stable enough to measure. That is a plant in volume production: levels 7 to 9 of Rolls-Royce’s Manufacturing Capability Readiness Levels, where the process is on the shop floor at volume with assured quality. There it is the right tool, and a good Black Belt is worth having in the room.

Most hard-technology companies are not there. A company building forty systems a year, a first plant, or a product still changing every quarter will never see a defect rate in parts per million; its statistics run out long before its sigma does. So the question changes. If the product is not made in volume, is reducing variation the measure that matters?

Not first. Before variation can be reduced there has to be a process that hits the target at all. For a new product the measure is hit rate: how many new things land inside their tolerance the first time, and how quickly they become repeatable.

Counting faults in parts per million only works if you make millions. That means a mature factory making the same thing in volume: levels 7 to 9 of Rolls-Royce’s Manufacturing Capability Readiness Levels. A company making a few dozen machines a year, or building its first plant, will never collect that much data.

For them the first question is simpler: does a new part come out right first time, and how quickly does it become routine? That is hit rate.

The board needs thousands of balls before the hill takes shape. Drop only forty and you see a scatter, not a curve. That is where most hard-technology companies stand: a few dozen systems a year, a first plant, a product still changing. They never drop enough balls to measure their sigma.

Their question comes earlier. When the first ball drops at a new target, does it land inside the lines? That is hit rate, and it matters before sigma does.

Kaizen versus step change: the kaizen fallacy

Kaizen, continuous improvement, narrows the curve a little each year. Each gain is smaller than the last, because every process has a floor: the machine, the material and the physics set the narrowest it can ever be. Improvement against that floor is worth doing. It never moves the curve to a place the process cannot reach.

Kaizen, the habit of small continuous improvements, makes the hill a little narrower each year. Each gain is smaller than the last, because the machine and the material set a limit on how narrow it can get.

Press Tighten the process and the hill narrows: fewer bounces, less spread. Every factory can do this for a while, a little each year. But each squeeze gives less than the last, because there are only so many pegs to take away.

Exhibit 2Optimise the past, create the future
  • Optimise the past. Each year the curve narrows on today’s target, by less each time, towards a floor (the dotted lines) set by the machine, the material and the physics.
  • Create the future. A new target needs a new process. It starts wide and has to be brought to capability.

Illustrative. Each row is a period of production; each curve is the spread of what the process makes.

A new product, a tolerance tighter than the machine can hold, a new material or a new site is a step, not a gain. It needs a new process, and a new process starts wide. The fallacy is to expect the old method to deliver the step: no amount of narrowing turns one curve into another.

Optimise the past; create the future. Six Sigma is a discipline for the first. The second is a different capability, and the companies that last do both.

A new product, a tighter tolerance, a new material or a new site is a different hill in a different place. Squeezing the old one never gets there. That needs a new process, which starts wide and has to be brought under control.

Improving the old process and creating the new one are different skills, and the companies that last have both.

No amount of squeezing moves the hill somewhere else. A new product is a new target: the lines are drawn in a new place, and the old board cannot reach them. You need a new board, and a new board starts wide.

Optimise the past; create the future. Six Sigma squeezes the old hill. Building the new board is a different skill.

The curve is chosen before the bid

The original Six Sigma knew this. Alongside the improvement projects it worked on design for manufacture: the decision that sets the width of the curve is made when the product is designed and the process is chosen, long before any project can improve it.

How wide the hill will be is mostly decided before production starts, when the product is designed and the process is chosen. Six Sigma, at its start, said the same.

The width of the hill is settled before the first ball drops, when someone chooses the board.

Exhibit 3Two processes, one tolerance
  • Process X. Cheaper per part, and it uses all of the tolerance: Cpk 1.0, about 2,700 parts per million outside the limits before any drift.Cheaper per part, and it uses all of the room the customer allows: about 2,700 rejects in every million parts, before it wanders at all.The cheaper board. Its hill fills the space between the lines: about 2,700 balls in every million go over.
  • Process Y. Dearer per part, and it uses a third of the tolerance: Cpk 2.0.Dearer per part, and it uses a third of the room: close to no rejects.The dearer board. Its hill sits well inside the lines, and almost nothing goes over.

Process X looks cheaper on the quotation. It uses up all of the tolerance, and the scrap, rework, rejections and recalls that follow cancel the saving. The cheaper process is the one with the lowest real cost, and choosing it takes judgement that no improvement project can supply afterwards. The same goes for tolerances. A bid is a commitment to meet every tolerance as written, so a supplier that needs a tolerance widened has to say so before it bids, not after the order is placed.

The cheaper process can be the dearer one. If it uses up the whole tolerance, the scrap, rework and returns that follow eat the saving. And a supplier that needs a looser tolerance has to ask before it bids, because a bid promises to meet the drawing as written.

Board X is cheaper to buy, and its hill fills the space between the lines, so the edges spill over. Board Y costs more, and its hill sits well inside. The spill from X is scrap, rework and returned parts, and it soon costs more than Y did. Choose the board before you sign: a bid is a promise to keep every ball inside the lines the customer drew.

Accuracy, repeatability and hit rate

Two words do most of the work. Accuracy is whether the average lands on the target. Repeatability is how tightly the results group. Six Sigma’s tools tighten the group and centre it, on a target the process already knows.

Accuracy is whether you hit the middle on average. Repeatability is whether your shots land close together. Six Sigma improves both, on a target you already know.

Now swap the board for a target. Accuracy is hitting the middle on average. Repeatability is the shots landing close together. Six Sigma coaches a shooter who already knows the target.

Exhibit 4Accuracy and repeatability
Accurate and repeatable
Repeatable, not accurate
Accurate on average, not repeatable
Neither

The capability to create capability asks a different question: when the target moves, how fast does the group arrive on it? Two plants can hold the same Cpk on today’s product. Give each four new targets. One lands most of them on the first attempt; the other needs several attempts at each. Their sigma is the same. Their hit rate on new things is not, and it is hit rate that decides whether the next product launches on time.

Kaipability cares about a third thing: how quickly a team hits a target it has never aimed at. Two factories can be equally good at today’s product and very different at the next one. That difference, the hit rate on new things, decides whether a new product launches on time.

Then move the target. Two shooters who were equally good on the old one are not equally good on a new one. One finds the middle with the first shots; the other needs many tries at each new target. That is the capability to create capability, and it decides whether the next product launches on time.

Exhibit 5Same Cpk today, different hit rate tomorrowEqually good today, not equally good at something newSame aim on the old target, different aim on a new one
Optimises today’s product
Today
New 1
New 2
New 3
New 4
1 of 4new targets hit first time
Has the capability to create capability
Today
New 1
New 2
New 3
New 4
3 of 4new targets hit first time

Illustrative. The dashed ring is the tolerance. Both plants group tightly on today’s product; what differs is the first attempt at something new.

What a Black Belt certifies

None of this is against the method. Data before opinion, the root cause before the fix, and a Control phase so that a gain is held are sound. So is Joseph Juran’s older test for whether people can be held responsible for quality: they must know what is required, be able to measure what they are getting, and have a process capable of meeting the requirement and the means to adjust it. The criticism is of what the belt is taken to prove.

  • A belt certifies training. No single body sets the standard that providers follow. ISO 18404 describes competencies but is voluntary, and certification is offered by professional societies, universities, consultancies and online providers, so rigour varies widely. A Black Belt shows that someone completed the training and, usually, led a project. It does not show that they have run a line, held a process capable at rate, or brought a new one up.
  • DMAIC improves a process that exists. It needs a stable process and a baseline to measure. Design for Six Sigma exists for new designs, but it is the DMAIC belt that is usually taught.
  • Savings are claimed project by project. Ask where a project’s saving shows in the accounts a year later, and whether the control plan is still followed on the floor.

The method itself is sound: measure first, find the root cause, then make sure the fix sticks. The trouble is the belt. A Black Belt shows that someone completed a training course, and courses vary widely. It does not show that they have run a production line, or brought a new one up.

None of this is a complaint about the method: measuring before arguing is good practice anywhere. The trouble is the belt. A Black Belt says someone learned to squeeze a hill. It does not say they have ever built a new board.

Optimise the past, create the future

The judgement is which one a programme needs, and when, before the money is spent. Kaipability works on both and obsesses over neither: sigma where volume earns it, a step where the future needs one. A plan that buys narrower curves for a product that has not yet hit its target spends money on the wrong problem. Capability is the same result the nth consecutive time, on second shift, at rate. The capability to create capability is doing that again on something new, sooner each time.

The skill is knowing which one a programme needs, and when, before the money is spent. Kaipability works on both: squeezing variation where volume justifies it, and building a new process where the future needs one. Capability is the same result the nth time in a row. The capability to create capability is doing that again on something new, sooner each time.

So the question for any plan is which one it is paying for: a narrower hill on the old board, or a new board for what comes next. Kaipability works on both and obsesses over neither. Capability is the same result the nth time in a row. The capability to create capability is building the next board, sooner each time.

The Kaipability “so what”

For an investor deciding whether to release capital, or an acquirer deciding whether to buy the capability, ask what the money is buying: a narrower curve on today’s product, or a new curve for tomorrow’s. The first is measured in sigma. The second is measured in hit rate, and a count of Black Belts says nothing about it. For an operator the same question decides the next capital request: a narrower curve on what the plant already makes, or a process for what it has to make next.

Questions

  • What is Six Sigma?

    Six Sigma is a quality method developed at Motorola in the 1980s. It reduces the variation in an existing process through data-led projects run in five phases, Define, Measure, Analyse, Improve and Control, by staff trained to belt grades. Its target is 3.4 defects per million opportunities.

  • What does a Six Sigma Black Belt certify, and is it worth it?

    That someone has completed Black Belt training and usually led an improvement project. No single body sets the standard providers follow, so rigour varies. It is worth having where a stable process needs a root cause found. It does not show the holder has held a process capable at rate the nth consecutive time, or brought a new one up.

  • What is the 1.5 sigma shift in Six Sigma?

    The 3.4 defects per million figure assumes the process average drifts by 1.5 standard deviations over time, so a six sigma process is scored as 4.5 sigma. Held centred, six sigma gives about 0.002 defects per million. The shift is a convention drawn from Motorola’s experience, not a statistical law.

  • Does Six Sigma work for low-volume manufacturing?

    In part. Its habits, data before opinion and the root cause before the fix, apply anywhere. Its statistics need volume: a defect rate in parts per million cannot be shown on a few hundred units a year. For low-volume and new products the measure that matters first is hit rate, how often a new part or process lands within tolerance first time, and how quickly it becomes repeatable.

  • What Cpk equals six sigma, and what is Cpk 1.33 in ppm?

    A Cpk of 2.0 is six sigma: six standard deviations from the process average to the nearer specification limit. A Cpk of 1.33 is four sigma. Held centred, a Cpk 1.33 process puts about 63 parts per million outside its limits; after the conventional 1.5 sigma drift, about 6,210.