Content uploaded by Roger Hoerl
Author content
All content in this area was uploaded by Roger Hoerl on Jun 24, 2015
Content may be subject to copyright.
THE EVOLUTION OF SIX SIGMA
Gerald J. Hahn and Necip Doganaksoy
(hahn@crd.ge.com and doganaksoy@crd.ge.com)
GE Corporate Research and Development
1 Research Circle
Niskayuna, NY 12309
Roger Hoerl
(Roger.Hoerl@corporate.ge.com)
GE Corporate Audit Staff
3135 Eastern Turnpike
Fairfield, CT 06431
KEY WORDS: Customer focus, design improvement, process improvement, statistical tools.
ABSTRACT
The Six Sigma approach to quality improvement has had a profound impact on the mode of operation, and
the profitability of the companies that have embraced it. Moreover, in some form or other, it is the wave of
the future for all organizations. We briefly review the Six Sigma approach and provide references for
further information. However, our major focus will be on a discussion of how Six Sigma has evolved--
especially its broadening from internal criteria, such as scrap reduction, to external ones, such as complete
customer satisfaction. This has led to a change in focus from mainly manufacturing considerations to
everything we do, and especially to product design and commercial operations. We also comment on some
lessons learned, and describe a specific application. Our comments are based directly upon our experience
in GE--but they have general relevance.
SIX SIGMA REVIEW
Basic Concept and Initial Implementation
Six Sigma is a disciplined and highly quantitative approach to improving product or process quality. The
original goal, implied in the Six Sigma definition, is the reduction of defects to no more than 3.4 per
million opportunities. This definition may, initially, divert us to questions, such as “What is a defect? ”,
“What is an opportunity? ”, “Is the implied 1.5 standard deviation average shift realistic? ”, and “Are all
products and processes created equal?” However, we have found that in real applications, the answers to
these questions, as well as the basic quality improvement goal, are generally clear. Almost invariably, this
calls for significant improvement over current levels of quality.
Acronyms
BB Black Belt
CTQ Critical to Quality Characteristic
1
DFR Design for Reliability
DFSS Design for Six Sigma
DMADV Define-Measure-Analyze-Design-Verify
DMAIC Define-Measure-Analyze-Improve-Control
DOE Design of Experiments
GB Green Belt
MBB Master Black Belt
SPC Statistical Process Control
Six Sigma was introduced at Motorola (see Harry (1994)) with the major goal of reducing defects of
manufactured electronics products. It has since been adopted and generalized by a number of companies,
such as Allied Signal and General Electric (GE). Some of the features that characterize Six Sigma are
•It is a tops-down, rather than a bottoms-up, approach, that is enthusiastically and unwaveringly, led by
such company CEOs as Larry Bossidy (Allied Signal) and Jack Welch (GE). Champions are appointed
from the ranks of the leaders in each business. They are responsible for ensuring the successful
implementation of Six Sigma in their own areas of influence.
•Both at the business and project level, Six Sigma leadership is, traditionally, the responsibility of
Master Black Belts (MBBs) and Black Belts (BBs). These roles involve full-time Six Sigma leadership
activities such as setting of quality objectives for the business and monitoring progress towards these
objectives, selection of Six Sigma projects, and mentoring and training project teams. Implementation
is the responsibility of the project team members (i.e., engineers, scientists, financial analysts,
information systems specialists, and so forth). They receive Green Belt (GB) level training from the
MBBs and BBs.
•It is a highly disciplined approach that typically involves the four stages Measure, Analyze, Improve,
and Control with an up front stage (Define) sometimes also added (DMAIC). In brief, these steps are
-Define (D): Define the problem to be solved, including customer impact and potential benefits.
-Measure (M): Identify the critical-to-quality characteristics (CTQs) of the product or service.
Verify measurement capability. Baseline the current defect rate and set goals for improvement.
-Analyze (A): Understand root causes of why defects occur; identify key process variables that
cause defects.
-Improve (I): Quantify influences of key process variables on the CTQs, identify acceptable limits
of these variables, and modify the process to stay within these limits, thereby reducing defect
levels in the CTQs.
-Control (C): Ensure that the modified process now keeps the key process variables within
acceptable limits, in order to maintain the gains long term.
•It is a highly data-oriented approach, as evidenced by such slogans as "We don't know what we don't
know" and "In God we trust--all else bring data." As a consequence, the DMAIC toolset is heavily
2
based on statistical tools and the statistical design of experiments (DOEs).
•It involves training everybody in the company in DMAIC, or modified concepts and tools--typically in
a 4-week program, spread out over 3 months. It is heavily hands-on project and bottom-line oriented.
In fact, one reason for the three-month training period is to allow students to apply the concepts and
tools to an important project. Upon completion, it is necessary to quantify and document the resulting
$ savings. Successful completion of Six Sigma projects is a requirement for Six Sigma certification.
The requirement for MBBs is typically successful mentoring of 20 projects and for GBs completion of
2 projects.
For further details, see Breyfogle (1999), the other articles in this issue of Quality Engineering, and
various other papers in the literature (Hahn, Hill, Hoerl and Zinkgraf (1999), Harry (1994), and Hoerl
(1998)).
The Results
The payoff from Six Sigma has been carefully scrutinized by Company accountants and auditors, and has
been impressive. This has included:
•Motorola - Almost a billion dollars of savings in three years, and a Malcolm Baldrige Award.
•Allied Signal: Over two billion dollars cumulative savings since it began Six Sigma.
•GE: Over a billion dollars of savings in 1998, and two billion anticipated for 1999. As the old saying
goes, "A billion here and a billion there, and pretty soon your talking some real money".
See Hahn, Hill, Hoerl and Zinkgraf (1999) for other specific examples. These results have helped company
leaders to maintain their fervor. For example, in announcing a new company initiative in electronic
commerce, GE CEO Jack Welch emphasized in an internal memo to all employees that Six Sigma is
absolutely critical to ensure the success of the new initiative.
Why Now?
We have seen various initiatives similar to Six Sigma come and go, during our careers. Although
moderately successful, none have had the all evasive impact of Six Sigma. So why is Six Sigma working
today?
One key factor has been the uncompromising commitment of top management. Little would have been
accomplished without this. But that support would only have been maintained if Six Sigma could
demonstrate success. And, in our opinion, this success has hinged heavily on technologies that have
emerged only recently. These include
3
•Automatic monitoring, accessibility to large (sometimes immense) databases,
•Ready availability to practitioners of powerful statistical tools (sometimes referred to as the
"democratization of statistics"), and
•Greatly improved communications capabilities that provide speedy transfer of information--including
videoconferencing, email, and, most recently, the Internet.
A highly disciplined, data-oriented approach, centered around the slogan "Show me the data--and do so by
no later than tomorrow morning--so I can factor into my decision,” has always made eminent sense. But, it
was not very realistic until recent technology made it so. The omnipresence of these technologies--together
with constantly increasing customer expectations and alternatives--is also why we feel that an approach
that, at least, resembles Six Sigma (although it may bear a different name in the future), is inevitable for all
organizations, including service enterprises--such as banks, hospitals, schools, and, even, local and national
government.
Most Six Sigma tools are not new and, in fact, have been around for many years. In our experience,
however, the impact of industrial training programs on specialized topics such as DOE, statistical process
control (SPC), regression and analysis of variance, dissipate quickly once the employees leave the
classroom and return to work. Six Sigma has been phenomenally successful in getting employees at all
levels of the organization to broadly use such tools long after completion of their formal training. In this
regard, one key to the success of the Six Sigma program is that there is an overall step by step roadmap
(DMAIC) that ties the tools together into an overall approach to improvement.
Finally, Six Sigma ties project benefit directly to bottom line results. Financial considerations play an
important role, starting with the early identification of project opportunities. The demonstrated success of
Six Sigma in impacting profitability gives it significant advantage over other, less direct, approaches.
THE EVOLUTION OF SIX SIGMA
Nothing in this world remains constant--even the best of things--and so it is with Six Sigma. As Six Sigma
moved from Motorola to other companies, and even as it has progressed within companies such as our own,
the emphasis has frequently changed. In this section, we describe various ways in which Six Sigma has
evolved. Although our comments here and elsewhere are based upon our experience within GE, we believe
they describe general trends among companies that have embraced Six Sigma.
Focus on Customer Satisfaction
The original motivation for Six Sigma at Motorola was centered around manufacturing improvement--and
4
this was also how Six Sigma was originally introduced in GE. There are various reasons. First, much of the
statistical methodologies of traditional quality control in general, and SPC in particular, has emphasized
manufacturing processes, and, thus many of the tools for Six Sigma were readily known. In addition,
payoffs, resulting, say, from reduction in scrap and rework of a manufacturing process, can often be easily
and, most importantly, speedily quantified. Such crisp and definitive cost reductions--preferably
documented within 3 months of the start of a project--were essential for building initial credibility for Six
Sigma.
However, important as manufacturing cost reductions may be, they often do not readily translate into
improvements that are transparent to the customer. GE management has come to realize this. In GE’s 1998
Annual Report, Jack Welch candidly stated “…as we celebrate our progress and count our financial gain,
we need to focus on the most powerful piece of learning we have been given in 1998, summarized perfectly
in the form of what most of our customers must be thinking, which is ‘When do I get the benefits of Six
Sigma. When does my company get to experience the GE I read about in the GE Annual Report?…What’s
the big event; and what did we miss?’” As a result, the central theme of Six Sigma in GE has rapidly
broadened from focusing principally on the manufacturing arena to encompassing all business operations,
and especially those that impact the customer. We will amplify further shortly.
Concentrate on Reducing Variability
Traditionally, Six Sigma quality could be achieved by some combination of improving mean performance,
and reducing variability. Often, however, it turned out that the former was easier to achieve than the latter--
and the major focus was, thus, on improving the mean. However, it soon became apparent that, in many
contexts--ranging from molding pellets into plastic parts to sending out bills for credit card payment--what
really impacted our customers was variability. Again quoting from GE’s 1998 Annual Report “The
problem is, as has been said, ‘the mean never happens,’ and the customer…is still seeing variances in when
(for example) deliveries actually occur--a heroic 4-day delivery time on one order with an awful 20-delay
on another, and no real consistency. The customers feel nothing…Their life hasn’t changed; their
profitability hasn’t increased one bit…Our challenge as we move toward 2000 is to turn our Company
vision ‘outside in,’ to measure the parameters of the customers’ needs and processes and work toward zero
variability in serving them. Variation is evil in any customer-touching process.” With such a stirring call to
action from top management, the focus of many Six Sigma projects in GE has rapidly shifted to reducing
variability. This has resulted in a whole range of concepts and tools, referred to as “Variance Based
Thinking.”
Design for Six Sigma
5
Deming told us long ago that the preponderance of product variability is in the system, rather than in the
process. This means that applying standard SPC approaches to a manufactured product is likely to lead to
only a fraction of the possible improvement. Instead, we need, initially, build quality into the design of our
products and processes. This can have significant payoff, both in direct cost reduction and customer
satisfaction because problems discovered in design, though, often difficult to identify, are, usually, easy to
fix. In contrast, problems found after the design has been frozen, and especially after significant amounts of
product have been built, although easy to identify, are often expensive to fix.
The DMAIC process outlined above is aimed mainly at reducing defect rates in existing products, services
and processes. The basic nature of designing something new requires a significantly different approach than
that required for “fixing” something that already exists. The need to adapt Six Sigma for design projects at
GE gave rise to Design for Six Sigma (DFSS). The objective of DFSS is to design products, services and
processes that are Six Sigma capable. A major goal is to minimize the occurrence of unpleasant last minute
surprises and hick-ups that are traditionally associated with introduction of new products, services and
processes. The basic principles of DFSS are
•Customer requirements: CTQs and other requirements for the new product, service or process are
defined at the customer level. This is achieved by disciplined use of customer research tools such as
the Quality Function Deployment.
•Requirements flow down: The customer requirements are gradually “flowed down” to requirements for
functional design, detailed design, and process control variables. This disciplined ensures that a holistic
systems view is maintained throughout the design process, and helps fight the urge to jump to finalize
design prematurely.
•Capability flow up: As requirements are flowed down, capability to meet these requirements is
continually assessed in light of relevant existing or new data. This permits early consideration of
potential tradeoffs as well as avoidance of otherwise predictable future surprises.
•Modeling: Both requirements flow down and capability flow up evolve on the knowledge of
relationships between customer requirements (Ys) and design elements (Xs). The models can be based
on physical fundamentals (e.g., kinetic models for a chemical reaction), simulation (e.g., discrete event
simulation model for a call center), empirical methods (e.g., response surface fit to data from a DOE)
or their mix.
The methodology used to implement DFSS is called DMADV, and involve the following five steps of a
disciplined data-oriented approach:
•Define (D): Identify the new product, service or process to be designed (or redesigned). Develop and
define a team charter, including scope, business case, milestones, resources and project plan. As an
example, we consider a team charged with designing a new thermoplastic resin for use in molding
exterior body panels of a car. The activities in the define phase are based on common sense and
6
constitute a major portion of any training program on project management. However, the
consequences of skipping or downplaying this phase can be quite severe. For instance, insufficient
definition of the project’s scope (e.g., “Does our project only involve coloring of the resin on a lab
scale equipment or translation to manufacturing as well?”) or its resource requirements from the rest
of the organization (e.g., molding equipment and test technicians, information systems specialists)
could seriously derail the whole project effort.
•Measure (M): Plan and conduct necessary research to understand the customer needs and associated
requirements. Translate these needs and requirements into measurable characteristics (CTQs). In our
example, it is desired that the molded parts are capable of attaining tight specifications on color
matching between adjacent car parts (e.g., doors and the fenders). Customer requirements on degree
of agreement between color of car parts is thoroughly explored and quantified. This might involve a
designed experiment whereby a panel of customers are asked to judge the difference between car
parts. The resulting observed color differences are then expressed quantitatively based on well
established spectrophotemetric measurements such as L (light-dark), a (red-green) and b (yellow-
blue). In addition to color, there are also other requirements on chemical properties and physical
performance characteristics of the resin.
•Analyze (A) Develop alternative concepts. Select the best fit concept for development into a high-level
design and predict the capability of the design to meet requirements. In the example, the key
determinants of color are the types and relative amounts of pigments, their addition to the product and
processing. At this phase, the various design options are considered and evaluated systematically.
This might involve the combined use of statistical experimentation and physical laws governing
behavior of pigments in thermoplastics in order to explore the relationships between the color of the
resin (as measured by L, a and b) and coloring agents.
•Design (D): Develop the detailed design. Evaluate the capability of the proposed design and develop
plans to pilot the new or redesigned product or service. In the example, the effect of variability in the
pigment amounts will be linked to capability of controlling the feeders on the manufacturing line. This
is an example of flowing down customer requirements (i.e., color match between adjacent car parts) to
manufacturing controls (i.e., tolerances on pigment feeders). If capability flow up suggests that the
existing feeders can not meet the requirements, their capability is addressed via a DMAIC project.
Likewise, the key properties of each pigment are eventually translated into specification on pigment
suppliers. In addition, a supplier DMAIC program is initiated in those areas that need be improved.
•Verify (V): Build and pilot a full-function, limited scale version of the new or redesigned
product/service. In the example, the full-scale production process is developed, start-up and validation
activities are performed, and the new or redesigned product/service is transitioned to the process
owners.
GE is now focusing on DFSS in its training either as an add-on module for those previously trained, or as
7
an alternative area of emphasis for new employees embarking on Six Sigma training.
Emphasize Design for Reliability
Reliability has been defined as "quality over the life" of a product or process. Again, initial
implementations of Six Sigma did not give much emphasis to reliability improvement. However, with the
new emphasis on Six Sigma improvements that will be felt directly by customers, it has begun to play an
increasingly more important role--since, after all, reliability is frequently that element of quality that our
customers experience. Thus, reliability has become a key CTQ for many products. Another element is the
evolution of long term service agreements under which the company takes on all maintenance and product
servicing responsibilities over the anticipated life of the product. These provide a great impetus to
improving long-term product reliability.
Just like quality, reliability has traditionally been considered, principally, after a product has been
manufactured--and, frequently, as a consequence of field failures. However, such a "ship and fix" approach,
in addition to being too expensive, is just not acceptable to our customers in today's highly competitive
environment. Thus, similar to the need for DFSS, and intimately related to it, is the goal of Design for
Reliability (DFR).
DFR requires a different toolset from, say, end of line quality improvement. This is due to several unique
aspects of reliability:
•One frequently need be concerned with accelerated testing (to ensure speedy responses),
•The available data often includes data on unfailed units (known as censored data),
•The normal distribution is usually not an appropriate model for analyzing time to failure data, and
•One is frequently most concerned with extreme distribution percentiles, such as the age by which 1%
of the product has failed (rather than, say, the mean or variability), or with estimating the probability of
survival to a specified age.
See Hahn, Doganaksoy and Meeker (1999) for further discussion. These topics, and appropriate software,
are stressed in current Six Sigma training for many engineering audiences.
Application to the Entire Business
Although most of the initial emphasis of Six Sigma was on quality improvement in manufacturing, it is
now being applied in key areas throughout the business, and beyond what would traditionally be considered
“quality”. Thus, a major focus of GE’s current Six Sigma efforts is to ensure that Six Sigma is being
applied to all business activities. Special emphasis is given to applying Six Sigma to commercial
8
transactions and to product servicing--both areas of great strategic importance to the company.
Commercial transactions encompasses GE’s non-manufacturing businesses, such as NBC, GE Information
Services, and especially GE Capital--as well as business transactions in GE’s more traditional
manufacturing businesses. Since GE Capital provides nearly 45% of GE's profitability, a tremendous
opportunity would be missed if the focus remained on only manufacturing.
At first glance, one may wonder how Six Sigma can apply to selling or insuring mortgages, bidding on
municipal bonds, insurance, or providing consumer credit. The key is an understanding of the fact that all
work occurs in inter-connected processes. These processes have outputs, which are our primary concern,
such as cycle time, profitability, accuracy, and so on. However, these outputs are a function of process
inputs, and what happens at key processing steps. For example, success in bidding for municipal bonds is
highly dependent on information inputs. These include knowledge of the financial risks and understanding
of competition, as well as the effectiveness of processing steps, such as estimation of default probabilities,
and preparation of the bid itself. If one is able to view both a manufacturing line and credit card collections
as processes with key inputs and processing steps to be improved, in order to improve outputs, the leap
from manufacturing to business applications becomes second nature. Similar considerations apply in
considering the time to fill an order, and the correctness of a billing process--just two examples of
commercial transactions in a traditional manufacturing business.
For illustration, a non-manufacturing Six Sigma project which involved collecting from delinquent credit
card customers is described here. In the consumer credit card business, losses from delinquent customers is
a huge expense, and one reason why credit card interest rates tend to be significantly higher than other
interest rates. In this case, the business had several branch offices trying to track down delinquent
customers for whom the phone number or address on file are not correct. When trying to trace such
customers, collectors have several options. One is to utilize other internal databases that might have a
correct address. For example, GE has numerous private label credit card accounts, such as Macy's, Home
Depot, and Exxon. The collector might also contact a credit bureau, which will provide addresses and
phone numbers of delinquent customers, for a price. There are also agencies that specialize in tracking
delinquent customers. Another option is to view a scanned image of the customer's original credit
application, in hopes that perhaps the number was transcribed incorrectly.
Of course, each of these methods has a cost associated with it, either a direct payment to an outside agency,
or at least the time for the collector doing the searching. An obvious question to ask when taking a Six
Sigma viewpoint is: Which of these methods is "best", on an absolute basis, as well as on a "value" basis,
considering costs? Also, one might wonder whether some of these techniques should be used in
conjunction, or whether there is so much overlap that once we have "struck out" with one method, we
9
might as well give up and write off the debt. In other words, at what point is it no longer financially prudent
to continue searching?
At the beginning of this Six Sigma project, every collector tended to have his or her own "method". That is,
some people would start with external databases, others would start with internal. Some collectors would
only search using one method, others would search with every method available before they gave up. These
differences among the collectors introduced extra variation in the process. Clearly, everybody couldn't be
"right"! Common sense would suggest that there must be an "optimal" method. But how could such a
method be found?
Historical data was interesting, but had some important deficiencies. For example, typically people used the
"cheapest" methods first, and only used expensive methods after they have "struck out" with these.
Therefore, the historic "hit rates" of the expensive methods were not particularly high, since they were only
used on "difficult" accounts. It was realized that this data had too much bias to accurately determine the
optimal approach. In such cases in manufacturing, one often utilizes DOEs. But this is a financial service,
and we all know that DOE doesn't apply to financial services, right?
In fact, a designed experiment was conducted, where each independent variable was one of the potential
tools, or databases used by collectors. Each tool was either used (+1 level), or not used (-1 level). The
dependent variables (Ys) were "hit rate" (the percent of the time that a person was found), cost, and cycle
time. A modified full 27 approach was used, with 100 accounts being searched for each combination of
tools. Obviously, a fractional design could have been used, but with the large number of collections, 128
combinations was not considered unreasonable. It did not matter how many times a delinquent customer
was found for a given combination, the customer was only considered "found" or "not found".
The analysis of the resulting data shed a great deal of light on collections. First, it turned out that the least
expensive tool was also nearly as effective, in terms of hit rate, as the most expensive ones. Secondly, it
was learned that there was less overlap between the tools than originally thought. Thus, those that could not
be found with two or three tools, could still often be located by yet another tool. Statistically, this showed
up in the analysis as a lack of interactions. The collectors expected there to be significant overlap between
tools, which would have shown up as antagonistic (negative) interaction for hit rate.
The experimental data led to an "optimal" search method, using several tools in sequence. If a person still
could not be located, the search was dropped, and the debt written off, since it would cost more money to
search than we could expect to collect. This procedure was taught to all collectors at all branches. The first
year financial benefits from collecting additional cash, and also avoiding writing off bad debt, was just
under $3 million dollars--gains that will continue to be enjoyed annually. Another benefit from this data-
10
based approach was that it provided a rational basis for deciding how many collectors were needed to
maximize collections. In this case, additional collectors were justified by the additional collections they
could produce. The cost of these additional positions was factored into the calculation of benefits given
above.
In addition to its Six Sigma initiative, GE has a service business initiative, targeted at such traditional
manufacturing businesses as GE Medical Systems, GE Aircraft Engines, GE Power Systems, and GE
Transportation Systems. This is in recognition of the growing strategic importance and potential
profitability of product service, versus manufacturing, and the competitive advantage that high-tech
servicing, facility management, remote diagnostics, and so on, can provide. Also, improvement of service
processes is much more difficult to "reverse engineer" by competition. This is because a motor or appliance
can be purchased and studied in excruciating detail. However, a remote diagnostics process, which is not
physically located in one place, and occurs through cooperation between diverse business functions, is
much more difficult for competition to copy. Six Sigma when applied to product servicing has some of the
same elements as Six Sigma applied to manufactured products and to commercial transactions--if again we
consider the service operation as an inter-connected process, with important outputs that are impacted by
specific inputs that we need understand and appropriately control.
New Roles for Master Black Belts and Black Belts
As indicated, the MBBs and BBs provide the overall leadership to the Six Sigma effort. Thus, an important
criterion for Six Sigma success is to select "top of the class" contenders as MBBs and BBs. But top of what
class? Originally, a prime consideration was technical knowledge and understanding. However, as the
training has developed and all have become more knowledgeable, the role of the technical expert is being
taken over by specialists, such as professional statisticians (perhaps, from a local university). Instead,
MBBs and BBs are being selected to a larger degree today based upon their management and
organizational skills. In fact, in GE they have been identified as a prime source--perhaps even the prime
source--for identifying and testing future top Company leaders. As a consequence, MBBs and BBs are
often expected to spend only a limited time in that role, before, hopefully, moving up in the organizational
ladder.
Getting Six Sigma into the Fabric of the Organization
The initial stages of a Six Sigma program generally involve top level attention on the program as well as
fairly major shifting of resources within the organization to create the infrastructure to support it (e.g.,
identification of top talent for the MBB and BB roles, training of project teams, initial round of projects,
etc). During this phase, the Six Sigma program becomes the main focus of project reviews, internal staff
11
meetings, company wide contests and celebrations, and external presentations. In today’s fast paced
business environment, no program can remain as something special for more than a few years. A successful
Six Sigma program becomes part of the fabric of doing things in the organization. In GE, this ingraining
process is referred to as making “Six Sigma the way we work.”
However, there are important issues that need be addressed to ensure lasting success of a Six Sigma
program. These include careful selection of projects, mentoring of project teams, adaptation of advanced
tools, setting and monitoring of quality objectives, refinement of Six Sigma to meet the needs of the ever
changing business environment, and targeted training. Therefore, companies that have initiated Six Sigma
programs will need to determine how to manage and maintain Six Sigma after the program reaches a level
of maturity. Although this topic has not yet attracted much discussion due to relatively small number of
companies that have adapted Six Sigma more than a few years ago, we expect that it will become a highly
debated issue as many more companies begin to adapt Six Sigma programs.
SOME LESSONS LEARNED
To proceed further in our discussion of the evolution of Six Sigma, we now consider some lessons learned.
Because it is always easier to identify areas for improvement than it is to do something about them, some of
these, unlike many of the changes described earlier, are still works in progress. However, they provide
some important clues as to the directions in which we expect Six Sigma to move in the future.
Factor in and Quantify Long-Term Payoffs
As previously indicated, in order to establish the credibility of Six Sigma, some immediate significant and
readily quantifiable payoffs were needed--thus leading to major emphasis on scrap and rework reduction.
Thus, initial Six Sigma projects were, rightfully selected, principally on the basis of anticipated short-term
payoffs. The emphasis on customer satisfaction, and DFSS have changed that and projects with payoffs in
the more distant future--and with riskier, but still potentially substantial payoffs--are becoming more
prominent. However, the Six Sigma initiative is still highly bottom-line oriented--as it should be. As a
consequence, in order to continue to encourage important long-term projects, we need develop convincing
measures of the anticipated payoff for such projects.
This is of particular concern in dealing with ensuring customer satisfaction, and in DFSS, especially for
reliability. In particular, how do we quantity customer satisfaction that is characterized by long-term repeat
purchases of new models of the current product (e.g., a wash machine), and for that matter, other products
from the same manufacturer (e.g., a refrigerator or a light bulb)? And in considering DFSS, how do we
quantify the gains realized from averted problems--especially if these problems are in product reliability
12
and might not come about until some time in the future (by when current management, if successful in the
short tem, will, likely, have been promoted)?
Consider Multiple CTQs
Closely related to the preceding concept is that of addressing multiple CTQs. Traditionally, an objective of
a Six Sigma project has been that of improving one, or a small number of CTQs. Often, however, all of the
important CTQs, and especially those that can not be immediately observed, such as product reliability, are
not taken into consideration. These can add to, or detract from, the total savings resulting from a Six Sigma
project. For example, in a program directed at reducing end of line scrap, we might, at the same time, be
enhancing reliability if we were engaged in identifying and eliminating the root causes for scrap. On the
other hand, if our "remedial action" was merely that of liberalizing our specification limits, we might be
succeeding at reducing scrap--but at the cost of worsening reliability.
Thus, in order to get a true measure of the true total impact of a Six Sigma project, we need consider and
quantify the impact on the project of all of the important CTQs.
Engage Suppliers
Many Six Sigma projects have found raw material or part vendors as creating the root cause of quality
problems. Therefore, to be successful, we need clearly involve suppliers in the Six Sigma effort. This
becomes increasingly important as we rely more heavily upon suppliers from throughout the world.
Moreover, accomplishing this can be especially difficult when we have limited leverage--as may be the
case when we provide only a small part of a supplier’s total business. In any case, we need clearly
communicate our needs, how we aim to measure conformance, what information we require from the
vendor (including information on process changes), and who will be responsible for what in demonstrating
that quality and reliability goals are being met.
Eliminate Differentiation Between Six Sigma and Non Six Sigma Projects
At the outset of the Six Sigma initiative, some important projects were clearly identified as being Six
Sigma projects and received special attention. At the same time, there was a tendency to give less attention
and resources to other important projects which were not designated as Six Sigma (perhaps, because the
payoffs were not immediately quantifiable or, perhaps, more uncertain--and not, necessarily, because they
were less critical). As Six Sigma becomes ingrained in the organization, this dichotomy tends to disappear.
The Green Belt trained employees are expected to utilize the Six Sigma approach and methodology in their
normal assignments.
13
Beware of Dogmatism
The Six Sigma toolset contains some powerful methods that have broad applicability. However, that does
not mean that each tool has applicability to each problem. For example, the DOE is an extremely useful
approach for getting meaningful information. However, in many applications, particularly in marketing and
commercial quality applications involving direct interface with customers, it may not be possible, or at least
prudent, to run a designed experiment.
Effective MBBs have become aware of the need of fitting the tools to the needs of their specific audience.
Thus, in the training programs, and in qualifying BB and GB projects for certification, they have learned
not to dogmatically insist on the use of a specific tool, but to use or adapt the available tools to best meet
the problem at hand. In addition, training programs are becoming more and more tailored to specific
audiences, as evidenced by special training for DFSS, DFR, and commercial applications--as well as
various added areas that we have not discussed, such as Six Sigma for Software Quality.
Utilize Broader Toolset
Standard Six Sigma training involves numerous tools as part of the process--see Hahn, Hill, Hoerl and
Zinkgraf (1999) for a typical BB and GB curriculum. Breyfogle (1999) and numerous books on statistics
provide extensive details on these tools. Most of these methods (e.g., process mapping, failure mode and
effects analyses (FMEA), process capability studies, measurement systems analyses, statistical thinking,
DOE, multiple regression) have demonstrated their value over time. A few, such as hypothesis testing and
analysis of variance (except as a tool for quantifying different sources of variability), would, in our opinion,
not suffer from a reduction in emphasis.
At the same time, other powerful tools have proven their worth, and warrant increased emphasis in future
training. These will become increasingly more important as we move beyond the simplest situations
(picking the proverbial “low lying fruit”) to more complex ones.
We have already indicated the importance of reliability as a CTQ for many customer-oriented Six Sigma
applications, and the need for greater focus on reducing variability--and the fact that these require the use
of some special tools. In this section, we elaborate on some further tools that warrant greater emphasis in
Six Sigma training.
Simulation Analyses. Simulation allows one to build a model of a process or system on a computer, and use
computer evaluations to assess the impact of alternative strategies. For example, simulation has
14
traditionally been used to model the building of a product in a factory and to assess the impact of adding
further equipment or personnel on the time to manufacture. Simulation can similarly be used to assess
business processes, such as the impact on customer waiting time of adding people to staff a product
telephone “hotline.” Also, simulation provides an attractive alternative to more formal statistical analyses
in, for example, assessing how large a sample is required to achieve a specified level of precision in a
market survey, or in a product life test.
Non-Normal Distributions. Many Six Sigma tools, at least implicitly, assume that the process is normally
distributed. For example, the concept of a Six Sigma process resulting in 3.4 defects per million
opportunities, implies (in addition to a long-term shift of 1.5 standard deviations in the mean) that the
random variability in the process follows a normal distribution. Many processes are, indeed, normally
distributed. In fact, there is a theoretical justification for normality based upon the statistical “central limit
theorem,” which shows that a process that is the sum of many small sources of variability results in
normality. However, others may be subject to only one or two important single factors--and, thus, can not
be represented as the sum of many small sources. In that case, there is no reason for one to expect the
process to be normally distributed. In addition, many processes are not symmetrically distributed, or can
not take on negative values. As previously indicated, product life is an example. Incorrectly assuming
normality can lead to false conclusions--especially when one is concerned with the “tails” of a distribution,
which is precisely what one is dealing with in seeking no more than 3.4 defects per million opportunities.
Fortunately, there are a wealth of alternative statistical models and tools that can be used when the
assumption of normality does not hold.
Advanced Modeling. Modeling of the relationship between the Xs and the Ys in a process play a very
important role in successful design projects. Some examples of models are kinetic models for a chemical
reaction based on physical fundamentals, discrete event simulation model for a service shop, and response
surface fit to data from a DOE. Such modeling effort dictates close partnering of diverse groups within the
organization. For example, design and manufacturing engineering often work hand in hand for construction
of the models, their validation and obtaining historical plant data to assess the impact of the new design.
DOEs and least squares regression analysis of historical data are the major Six Sigma tool used for
developing such relationships empirically.
15
Also, other advanced analytical tools that deal with relationships are often useful. For example, a method
known as Classification and Regression Trees (CART) can help segment populations. This is frequently
required in commercial applications, especially when there is a large database. A typical example of the use
of CART arises in identifying those individuals in a large population to solicit in marketing a product. In
this case, CART would use data on a past sample of purchasers and non-purchasers to establish sub-groups
in the population with varying “marketing success probabilities,” based upon known characteristics, such as
marital status, geography, and past purchases.
Advanced DOE Concepts. Standard Six Sigma training has focused on factorial, fractional factorial, and,
sometimes, response surface, designs. DOE is often assumed to be a “one-shot” investigation. In
practice, however, experimentation is a sequential process, involving a series of trials as the learning
process evolves, especially when one is searching for an optimum. Thus, sequential experimentation
(see Box, Hunter and Hunter (1978) and Box(1999)), merits more consideration. So do “nested”
experiments” in which one variable is nested within another. A typical example arises when one is
assessing the impact of machines and operators on process variability, and when each operator is
using a different machine (see, for example, Box (1998)).
Survey Sampling Tools. Survey sampling deals with sampling from existing, frequently human,
populations. This is especially relevant in obtaining customer feedback. In sampling human populations we
are especially concerned with ensuring a random--or, at least, representative--sample. Key questions often
revolve around the required sample size and how to minimize (and handle) non-respondents so as to
minimally bias the results. Survey sampling approaches complement, for human and other existing
populations (e.g., invoices), DOE for processes or products.
Added Graphical Tools. Standard Six Sigma training, rightfully, emphasizes the use of graphical tools,
such as box plots, to allow one to extract information from data without becoming involved in, relatively
unfamiliar, statistical analyses--and to supplement such analyses. We advocate even greater reliance on
graphical tools--for example, in analyzing the results of a DOE and in plotting data (or residuals from a
regression analysis) over time.
We do not wish to suggest, by the preceding discussion, that Six Sigma training be turned into a course in
advanced statistics. However, we feel that various “non-standard” tools, that are proving their worth in Six
Sigma applications, will have increased relevance in future applications (such as CART analyses for
commercial transactions). Thus, technically oriented MBBs and BBs should be knowledgeable in them and
their applicability, and others should have, at least, some minimum level of awareness.
CONCLUDING REMARKS
16
Six Sigma continues to thrive, as evidenced by the success of those companies that have embraced it with
fervor, and the many other organizations that are expressing interest. In our opinion, the key elements of
success are the combination of a highly disciplined approach with one that is intensely data driven, the
ready accessibility of appropriate technical tools to leverage the concepts, and the continuing
uncompromising commitment from top management.
Six Sigma is a highly dynamic approach. We have tried to describe its evolution. A significant element is
the broadening of the general approach. This has resulted in greater emphasis on customer satisfaction and
reducing variability in performance. It has led to extending the use of Six Sigma tools from mainly short-
term manufacturing situations to product design and commercial applications. In turn, this has resulted in a
change in the role of those that lead Six Sigma, the need for added tools, and various other developments
discussed in this paper. This evolution will continue and, perhaps, even accelerate over time. However, we
are confident that, irrespective of the specifics, future developments will build on the solid foundation that
the Six Sigma approach has already provided.
REFERENCES
Box, G. (1998). Quality Quandaries--Multiple Sources of Variation: Variance Components, Quality
Engineering, 11, 171-174.
Box, G. (1999). Quality Quandaries--The Invention of the Composite Design, Quality Engineering, 12,
119-122.
Box, G.E.P., Hunter, W.G., and Hunter, J.S. (1978), Statistics for Experimenters, John Wiley & Sons, New
York, N. Y.
Breyfogle, F.W. (1999), Implementing Six Sigma: Smarter Solutions using Statistical Methods, John Wiley
& Sons, Inc., New York, NY
Hahn, G.J., Doganaksoy, N., and Meeker, W.Q. (1999), Reliability Improvement: Issues and Tools,
Quality Progress, May, 133-139.
Hahn, G.J., Hill, W.J., Hoerl, R.W, and Zinkgraf, S.A. (1999), The Impact of Six Sigma Improvement--A
Glimpse into the Future of Statistics, The American Statistician, 53, 3, August., 208-215.
Harry, M. (1994), The Vision of Six Sigma. Roadmap for a Breakthrough, Sigma Publishing Company,
Phoenix, AZ.
17
Hoerl, R.W. (1998), Six Sigma and the Future of the Quality Profession, Quality Progress, June, 35-42.
----------------------------------------
Gerald J. Hahn is Manager, Applied Statistics Program, GE Corporate Research and Development. He is a
Fellow of the American Statistical Association and of the American Society for Quality, and a GE Coolidge
Fellow.
Necip Doganaksoy is a Quality Program Manager at GE Corporate Research and Development. He is a
Fellow of the American Statistical Association.
Roger Hoerl is the Quality Leader of the GE Corporate Audit Staff. He is a Fellow of the American Society
for Quality, and the American Statistical Association.
18
