Six Sigma and Quality Control: How the Approach Builds Quality In

Every defect is paid for twice: once when the process makes it, and again when someone finds it, fixes it or apologises for it. Six Sigma builds quality into a process by putting a price and a number on that waste, tying the number to what the customer actually asked for, and then fixing the process so the number stays down. This article follows one back-office process through all of it, with the sums shown.

Why does a defect cost far more than the item it spoils?

The expensive part of a defect is rarely the material. A faulty part has to be scrapped or reworked, but before that it has to be found, moved, logged and re-checked, and if it escapes the factory it triggers a return, a replacement, a courier and a conversation with an unhappy customer. Quality professionals group these costs under the heading cost of poor quality, and the usual four-way split is the quickest way to see where the money goes.

Cost group	What it covers	Typical examples
Prevention	Spending to stop defects being made	Operator training, preventive maintenance, process design reviews, error-proofed fixtures
Appraisal	Spending to find defects	Inspection, testing, calibration, audits, sampling plans
Internal failure	Defects caught before the customer sees them	Scrap, rework, re-inspection, downtime, retesting
External failure	Defects the customer finds	Returns, warranty claims, replacements, complaint handling, lost orders

A worked month in a small plant

Take an illustrative plant that ships 40,000 units a month. Six per cent of units, 2,400 in all, fail somewhere along the line: 900 are scrapped at $18 each and 1,500 are reworked at $9 of labour each. A further 160 units slip past inspection and come back from customers at an average handling cost of $210 each. The plant also spends $9,000 a month on inspection and $3,000 on training and maintenance.

Cost group	Monthly cost	How it is built up
Prevention	$3,000	Training and maintenance
Appraisal	$9,000	Inspection labour
Internal failure	$29,700	900 x $18 scrap plus 1,500 x $9 rework
External failure	$33,600	160 returns x $210
Total	$75,300	Failure costs are $63,300, which is 84% of it

Only $3,000 of the $75,300 is spent on stopping defects, and $63,300 goes on dealing with the ones already made. Now suppose a Six Sigma project adds $4,000 a month of prevention, a better fixture and a day of operator training, and cuts the defect rate from 6% to 2%. Scrap falls to 300 units ($5,400), rework to 500 units ($4,500) and returns to about 50 ($10,500). The new total is $7,000 prevention, $9,000 appraisal, $9,900 internal failure and $10,500 external failure, which is $36,400. The plant saves $38,900 a month after paying for the extra prevention, and none of that saving needed a bigger inspection team.

The figures are invented for the illustration, but the shape is the one real projects find: failure costs dwarf prevention, so moving money towards prevention pays. ASQ's own Certified Quality Improvement Associate body of knowledge names cost of poor quality as an improvement technique candidates must be able to select and use, which is why it appears again in the exam section below.

How does Six Sigma turn quality into a number you can track?

Six Sigma counts defects instead of describing quality in adjectives. Three terms carry most of the weight. A unit is whatever the customer receives: a part, a loan decision, an invoice. An opportunity is a distinct way that unit could be wrong. A defect is a failure at one of those opportunities. A unit with two defects counts as two defects, which is what lets a rate be compared across processes of different size.

Defects per million opportunities, step by step

Switch to a service example that the rest of this article will keep using: the back office that processes consumer loan applications. In one month it handles 1,200 applications. Each application has five places where it can go wrong: the income figure, the signature, the identity document, the product code and the consent form. The review finds 87 defects in total.

1. Opportunities. 1,200 applications x 5 checks = 6,000 opportunities.
2. Defect rate. 87 defects / 6,000 opportunities = 0.0145.
3. DPMO. 0.0145 x 1,000,000 = 14,500 defects per million opportunities.
4. Defects per unit. 87 / 1,200 = 0.0725, so an application has about a 7% chance of arriving with at least one defect (the Poisson approximation gives a first-pass yield of e to the power of -0.0725 = 93.0%).

The sigma level follows from the DPMO. Convention adds a 1.5 sigma shift to allow for the way processes drift over time, so the sigma level is the normal-curve z-value for the yield plus 1.5. For 14,500 DPMO that is z = 2.18, and the sigma level is 3.68.

Sigma level	Defects per million opportunities	Yield
2	308,538	69.15%
3	66,807	93.32%
4	6,210	99.379%
5	233	99.9767%
6	3.4	99.99966%

Why 99% accurate can still be a poor process

A 99% yield sounds comfortable, yet it equals 10,000 DPMO, a sigma level near 3.8. A team processing 10,000 applications a month with a 99% accuracy rate sends out 100 flawed decisions every month, and if each of the five checks is 99% right on its own, only about 95% of applications pass every check. The sigma scale exists because ordinary percentages hide that: the gap between 99% and 99.99966% is the gap between a hundred errors a month and fewer than one a year for the same workload.

One caution on reading the table: the 1.5 sigma shift is a convention, not a law of nature. Some organisations report short-term sigma without it, so always ask which convention a quoted sigma level uses before comparing two figures.

How do customer needs become requirements a process can be measured against?

A defect only means something against a requirement, and customers rarely state requirements in measurable form. Six Sigma bridges the gap by capturing the voice of the customer, the raw wording from surveys, complaints, warranty claims and interviews, and translating it into critical to quality characteristics, usually shortened to CTQs. A CTQ is a requirement with a measure and a limit attached.

The translation runs in three moves: collect what customers say, ask what drives that statement, and then write the driver as something you can count or time. In the loan back office it looks like this.

What the customer says	What drives it	CTQ with a limit
"I want to know quickly whether I can have the loan."	Time from submission to decision	Decision issued within 72 hours of a complete application
"Stop asking me for the same document again."	Applications returned for missing or unreadable items	No more than 1 in 100 applications sent back to the customer
"The figures in my offer were wrong."	Accuracy of income, product code and rate fields	Every offer matches the approved application on all fields
"Nobody told me what was happening."	Status messages at each stage	Confirmation within one working day, update at each stage change

Three habits keep a CTQ list honest. First, split customers: the applicant, the loan officer who receives the file and the regulator who audits it all have requirements, and an internal customer's CTQ is as valid as an external one's. Second, write every CTQ with a unit and a limit, because "fast" cannot be measured and "72 hours" can. Third, check the CTQs with the customer before any measuring starts, since a project that measures the wrong characteristic perfectly still fails.

The same list also explains the five opportunities in the DPMO example above. The signature and consent checks exist because a missing one forces a return to the applicant, which breaks the second CTQ. A defect definition built from CTQs keeps the count tied to what customers care about rather than to what is easy to count.

Can the process actually meet the limit it has been given?

Process capability compares two things: what the customer allows (the specification) and what the process actually delivers (its natural spread). Six Sigma calls these the voice of the customer and the voice of the process. A process is capable when the spread fits comfortably inside the limit, with room to spare for drift.

Return to the first CTQ, a decision within 72 hours. Suppose the measured decision times average 58 hours with a standard deviation of 10 hours, and 72 hours is the upper limit. The limit sits (72 - 58) / 10 = 1.4 standard deviations above the average, so about 8.1% of applications breach it.

Change tried	Average	Std deviation	Limit in std deviations	Late applications
Today	58 h	10 h	1.40	8.1%
Same-day queue triage moves the average	52 h	10 h	2.00	2.3%
Standard checklist narrows the spread as well	52 h	8 h	2.50	0.6%

The table shows the two levers every capability study offers: move the average, or shrink the spread. Moving the average is usually quicker because it is a matter of adjustment, while shrinking the spread needs the sources of variation understood and removed. Capability indices such as Cpk express the same arithmetic as one number. For a one-sided limit like this one, Cpk is the distance to the limit in standard deviations divided by three, so the three rows above are 0.47, 0.67 and 0.83. Many customers want 1.33 or more, which is roughly four standard deviations of room.

These are short-term figures from a stable process. If the process is not stable, if the average wanders or the spread jumps from week to week, the capability number describes nothing, which is why the control step later in this article matters. Ask first whether the process is stable, then whether it is capable.

Which defects deserve the team's effort first?

A process rarely fails in one way, and fixing the rarest failure first wastes the project. Six Sigma teams sort defects by type and count before they hunt for causes, using a Pareto chart: bars in descending order with a cumulative line. The principle behind it, that a few causes produce most of the effect, is a pattern rather than a rule, but it appears so often that checking for it is standard practice.

The 87 defects from the loan example break down as follows.

Defect type	Count	Share	Cumulative
Wrong income figure	31	35.6%	35.6%
Missing signature	24	27.6%	63.2%
Identity document unreadable	14	16.1%	79.3%
Wrong product code	11	12.6%	92.0%
Missing consent form	7	8.0%	100.0%

Two of five defect types account for 63% of the problem. Removing both would take the count from 87 to 32, the DPMO from 14,500 to 5,333 and the sigma level from 3.68 to 4.05, with no work at all on the other three.

From the biggest bar to the cause

Once the top bars are chosen, the team asks why they exist. A fishbone diagram sorts possible causes by category, and the 5 Whys follows one chain to the bottom. For the income errors the chain might run: the figure is wrong because it was keyed by hand; it was keyed by hand because the payslip comes as a scan; the scan is read by a person because the form has no structured income field; the form has none because it was copied from an older product. The fix is a structured field with a range check, not a reminder to be careful.

Notice what the chain does not end on: a person. When a root cause turns out to be somebody's carelessness, the analysis usually stopped a step too soon. Six Sigma's working assumption is that a process producing defects is built to produce them.

How does the improvement stay in place after the project team leaves?

Most of the gain from an improvement project is lost in the months after it ends, and the cause is plain: the people who understood the fix move on, and the old habits return. The last stage of the Six Sigma method, control, is the work of making the new level the default.

The elephant on a small helmet in the picture is a fair image of a quality result that depends on one point of support: somebody remembering to check. Control plans, charts and mistake-proofing widen that base.

A control plan puts the fix in writing

A control plan is a short document that names, for every critical characteristic, how it is measured, how often, by whom, and what happens when the reading is bad. A usable one answers six questions.

Element	What it states	Loan back-office example
Characteristic	The CTQ or process input being held	Income field accuracy
Specification	The limit and target	Matches the payslip; range check 0 to 20,000 a month
Measurement method	The check and who runs it	Weekly audit of 50 files by a team lead
Frequency and sample	How much, how often	50 files every Friday, drawn at random
Reaction plan	What to do when it fails	Two bad files in a sample trigger a same-day review of the form logic
Owner	One named person	Operations manager, not the project team

The owner row decides whether the plan survives. A control plan owned by the project team lapses when the project closes; one owned by the process manager sits inside the weekly routine.

Control charts say when to react and when to leave things alone

A control chart plots a measure over time against a centre line and limits that describe how the process behaves when only its everyday, common-cause variation is acting. A point outside the limits, or a long run of points on one side of the centre line (many organisations use eight or nine), signals that something special has changed. A point inside the limits signals nothing, however much it disappoints a manager.

That distinction protects the gain from two opposite errors. Reacting to every wobble, which Deming called tampering, adds variation. Ignoring a real shift lets the defect rate drift back up unseen. The weekly count of returned applications, plotted on a control chart, tells the owner which of the two situations they are in. The calculation of the limits is a topic of its own; what matters for control is reading the signal and acting on it.

Mistake-proofing removes the need to remember

The most durable controls are the ones that make an error impossible or immediately visible. This is poka-yoke, the Japanese term for mistake-proofing associated with Shigeo Shingo at Toyota. There are two forms:

• Prevention. The process cannot move forward with the error present. A form that will not submit without a signature removes the 24 missing-signature defects at the source, and a drop-down for the product code makes the 11 wrong-code defects impossible to type.
• Detection. The error is caught at once, while it is cheap. A range check that flags an income entry of 200,000 when the payslip says 2,000 does not stop a slip but exposes it in seconds rather than at the audit.

Prevention beats detection because it costs nothing per transaction and depends on nobody. Everyday examples include a fuel nozzle that fits only the right tank and a plug that goes in one way round. Reinforcing the controls with a standard work instruction, a short training record for new starters and a periodic audit completes the hold.

Where does this show up in the ASQ CQIA exam?

The Certified Quality Improvement Associate exam from ASQ is the foundation-level credential that tests these ideas at the level of an informed team member rather than a project leader. The ideas in this article map onto its body of knowledge directly, as the syllabus topics page lists them.

• Improvement domain (40 questions). Six Sigma concepts and the five DMAIC phases at the Understand level; cost of poor quality as an improvement technique at the Apply level; Pareto charts, histograms and control charts as improvement tools; the 5 Whys and fishbone diagram for root cause analysis; poka-yoke among the lean tools.
• Quality Basics (30 questions). Common and special cause variation, the SIPOC view of a process, and the difference between quality assurance, quality control and continuous improvement.
• Customer Relationship (7 questions). Voice of the customer, complaint handling and quality function deployment.

The remaining domains, Team Basics with 16 questions and Supplier Relationship with 7, are about people and suppliers rather than process numbers. The facts below follow ASQ's CQIA page.

CQIA exam detail	Figure
Exam code	CQIA
Vendor	ASQ
Questions	110 (ASQ counts 100 as scored and 10 as unscored)
Duration	210 minutes total appointment time, 198 minutes exam time
Exam fee	ASQ members $334, non-members $434, retakes $234
Passing score	550/750
Open book	Yes, with bound reference material

Cost-of-poor-quality, Pareto and control-chart questions are easiest to rehearse in the real multiple-choice style, and the CQIA practice exam gives you that before you book. Fees, delivery options and testing windows are on ASQ's own CQIA page, which is the place to confirm them before applying.

What does a candidate need to be able to do with these numbers?

At CQIA level the exam rarely asks for a DPMO sum from scratch, but it expects a candidate to read figures like those above and decide what they imply. Five abilities cover most of the questions of this kind.

1. Read a Pareto chart. Pick the category that should be attacked first and say why the smallest bar is not it.
2. Classify a cost. Given a scenario such as a retest, an audit or a warranty replacement, name the cost group it belongs to.
3. Tell variation apart. Decide whether a point on a control chart is common cause, which calls for no reaction, or special cause, which calls for investigation.
4. Turn a complaint into a requirement. Recognise a voice-of-the-customer statement and the measurable characteristic that sits behind it.
5. Choose prevention over detection. When two fixes are offered, prefer the one that stops the error being made.

A short routine works well: take one process you know, write its defect types and count them, sort them into a Pareto order, assign each defect to a cost group, and write one control-plan row for the largest. That single exercise touches every item on the list above, and it makes the vocabulary stick in a way that reading definitions does not.

Six Sigma does not guarantee quality by decree; it makes quality measurable, finds the defects that matter most, removes their causes and puts the removal under watch. The number in the middle of this article, a loan process moving from 3.68 to 4.05 sigma by dealing with two of five defect types, is how the method looks in practice.

Frequently Asked Questions

Does the CQIA exam cover Six Sigma?

Yes. ASQ's Certified Quality Improvement Associate body of knowledge has a 40-question Improvement domain covering Six Sigma concepts and the DMAIC phases, cost of poor quality, Pareto and control charts, root cause analysis and poka-yoke. Voice of the customer is tested in the Customer Relationship domain.

How many questions and how much time does the CQIA exam give you?

The computer-delivered CQIA exam has 110 multiple-choice questions, of which ASQ scores 100. The total appointment time is 210 minutes and the exam itself runs 198 minutes. The passing score is 550 on a scale of 750, so pacing across all five domains matters more than speed on any one.

How much does the ASQ CQIA exam cost?

The initial exam fee is $334 for ASQ members and $434 for non-members, and a retake costs $234. ASQ sets these prices and can change them, so check the current figure on ASQ's CQIA page before you apply or pay for a sitting.

Can you bring notes to the CQIA exam?

ASQ exams are open book, but the rules on what you may bring are strict. Reference material, including your own notes, must be bound, and a test centre administrator inspects it before you enter. Loose pages and unfastened staples are refused, so bind your notes properly well ahead of the day.

What does ASQ require before you can sit the CQIA?

ASQ's CQIA page asks for a full-time, paid role, together with either two years of work experience or an associate degree or two years of equivalent higher education. It names no Six Sigma belt as a prerequisite, so the credential is open to people who have not yet joined a Six Sigma project.

Select ratingPoorOkayGoodGreatAwesome

Rating: 5 / 5 (76 votes)