- The Expansion to Lean Six Sigma
- Macro Societal Shifts Since 1987
- Current State of the Art
- The Limitations of Lean Six Sigma 1.3
- A New Paradigm Is Needed
- References
This chapter is from the book
This chapter is from the book
This chapter is from the book
The Limitations of Lean Six Sigma 1.3
Lean Six Sigma 1.3 offers many commendable improvements. For example, it covers a diverse array of application areas, from Internet commerce and other high-tech industries to healthcare, finance, and, of course, manufacturing. It has incorporated key Lean principles from the Toyota Production System, such as line of sight and 5S, to drive improvement even before data is collected. Furthermore, research has documented a clearer, more synergistic relationship between Lean Six Sigma and disruptive innovation, and also demonstrated how to use Design for Six Sigma projects to take innovative concepts to market. Despite these advantages, we feel that Lean Six Sigma 1.3 still needs to overcome the following limitations:
Still not appropriate for all problems
Does not incorporate routine problem solving
Does not provide a complete quality management system
Cannot efficiently handle large, complex, and unstructured problems
Does not take advantage of Big Data analytics
Does not address modern risk management issues
In this section, we highlight the importance of each limitation in greater detail. In our view, minor adjustments to Lean Six Sigma 1.3 will not address these limitations. Instead, a new paradigm is needed. Kuhn (1962) noted that the need for a new paradigm becomes apparent when the list of problems that existing paradigms cannot solve becomes too large to ignore. We feel that Lean Six Sigma 1.3 is now at this place.
Still Not Appropriate for All Problems
Six Sigma, not even in the form of Lean Six Sigma 1.3, is the most appropriate approach for all projects. For example, the second author of this book (Hoerl) was asked years ago to help a computer scientist with his Six Sigma project at GE. When Hoerl asked him about the project, he stated that it involved installing an Oracle database. Hoerl asked if he knew how to install an Oracle database, and he replied yes. Hoerl asked if he had already done this successfully, and the computer scientist again replied yes. Now with a puzzled look on his face, Hoerl asked what problem required a solution. The computer scientist replied that there was no problem to be solved, but his boss had told him to use Six Sigma on this installation, so this is what he was going to do.
This is a classic case of a “solution known” problem (Hoerl and Snee, 2013). We have a problem, but the solution is already known. This does not necessarily mean that the solution is easy to implement—properly installing databases is not trivial. However, there is no need to analyze data to search for a solution. We just need to ensure that the people doing the work have the right skills and experience, and perhaps procedures, to properly implement the known solution. The question of whether the solution is known or unknown is a key consideration in choosing a methodology, as we see in the next chapter.
The key point, we hope, is clear. Six Sigma was not needed, and perhaps not even helpful, for that installation. Of course, some type of formal project management system, and possibly database protocols, were needed to ensure success. But Six Sigma was not needed. Over the years, we authors have both had numerous similar conversations with people who were trying to force-fit Six Sigma where it was not needed—and sometimes where it was not appropriate. For example, as in this case, Six Sigma is helpful only for “solution unknown” problems. Simpler methods can address more straightforward problems. These include Work-Out, a team problem-solving method developed at GE (Hoerl, 2008), so-called “Nike projects” (Just Do It!), and “Is–Is Not” analysis (Kepner and Tregoe, 2013), to name just a few. We discuss each of these techniques within the context of a holistic approach to improvement in subsequent chapters.
Of course, integrating Lean into Six Sigma helps avoid force-fitting Six Sigma because Lean might be an appropriate methodology for a given problem when Six Sigma is not. For example, Lean has proven principles that provide excellent guidance on “solution known” problems (Hoerl and Snee, 2013). However, just as Six Sigma is not appropriate for all problems, Lean also is not appropriate for all problems. In short, any time we select the problem-solving methodology before we have clearly documented the problem, we are prone to force-fitting. Shouldn’t we learn about the problem first and only then determine the best approach to find a solution? After all, no one would continue to see a physician who recommends treatment before learning about the patient’s condition.
Does Not Incorporate Routine Problem Solving
Lean Six Sigma 1.3 does not incorporate routine problem solving. Suppose that a manufacturing line begins leaking oil at 3:30 AM. Clearly, this is not the time to put together a Six Sigma team to gather data and study the problem for a few months; someone needs to promptly stop the leak! Similarly, if someone notices mislabeled medication in a pharmacy, we wouldn’t want the pharmacy to put together a Six Sigma team; it needs to identify and remove the mislabeled medication immediately, before anyone receives it.
By “routine problem solving,” we mean the normal, day-to-day problem solving that occurs in all organizations, typically in real time. Of course, some people and some organizations are particularly good at it, and others aren’t. The ideal is to have each employee well trained in how to approach routine problems, diagnose root causes, and identify and test solutions. Employees should also understand when to call for help. Most of the problems faced in the workplace, and even in our private lives, can be solved in a short amount of time with no data or minimal data. In the case of the leaking oil, you would just follow the oil to find the source of the leak.
Of course, some problems are not easily solved on a routine basis. For example, suppose that this is the fifth time this year that one of the machines in the plant has begun leaking oil. Why is this problem reoccurring? Is the fundamental root cause the oil itself, the equipment, the way we are operating the equipment, the way we are maintaining the equipment, or something else? To solve this higher-level problem, a team and some formal methodology (perhaps Six Sigma) is likely needed. The key point we are making is that routine problem solving is an important aspect of continuous improvement; however, we typically do not need Lean Six Sigma, nor is there time to conduct a lengthy project. Immediate solutions are needed.
Not a Complete Quality Management System
Fundamentally, Lean Six Sigma 1.3 is a project-based methodology for driving improvement, but it is not a complete quality management system. That is, it does not replace ISO 9000 quality systems or provide the same breadth as national quality awards, such as the Malcolm Baldrige National Quality Award (MBNQA) in the United States. For example, individual Six Sigma projects might lead to calibrating measurement equipment in a lab during the Measure phase, but they would not provide an overall laboratory calibration system. Similarly, one element of the MBNQA addresses strategic planning; however, we do not advise organizations to develop their strategic plan through Six Sigma projects. Of course, we hope that continuous improvement and enabling methods such as Lean Six Sigma are key elements of the strategic plan.
We could give many other examples here. Our point is simply that an organization’s overall quality system should be developed in a top-down manner, based on its philosophy and strategic plan, to help meet business objectives. Six Sigma projects, on the other hand, are narrowly focused on specific problems that have been identified and that typically can be solved in roughly three to six months. Conversely, an overall quality management system should not be developed from the bottom up, based on a set of individual projects that were chosen for other purposes. If this is the case, then, by definition, the quality management system is not strategic in nature.
Before describing a recommended paradigm to address the limitations of Lean Six Sigma 1.3, we elaborate further on a few related phenomenon that we mentioned briefly previously: the issue of large, complex, unstructured problems; the emergence of Big Data analytics; and the increased importance of risk management in today’s world.
Inefficient at Handling Large, Complex, and Unstructured Problems
We noted earlier that, since 1987, there has been a growing awareness that some problems are too large, complex, and unstructured to be solved with traditional problem-solving methods (including Lean Six Sigma). As noted by Hoerl and Snee (2017), applications in such areas as genomics, public policy, and national security often present significant challenges, even in terms of precisely defining the specific problem to be solved. For example, in obtaining and utilizing data to protect national security, we could perhaps develop an excellent system relative to surveillance and threat identification, but it would essentially result in a police state limiting privacy and individual rights. Few people would consider this a successful or desirable system.
Similarly, the system currently in place for approving new pharmaceuticals in the United States involves a series of clinical trials and analyses guided by significant subject matter knowledge, such as identifying likely drug interactions. No single experimental design or statistical analysis results in a new approved pharmaceutical. Furthermore, the system must balance the need for public safety with the urgent need for new medications to combat emerging diseases such as Ebola or Zika. This problem is complex.
Such problems are unique from most others that can be solved through routine problem solving, or even through Lean Six Sigma. In the following subsections, we briefly discuss some of the important attributes of these types of problems:
Too Large to Tackle with One Method
In terms of size, the problem is simply too large to be solved with any one methodology. Several tools, and perhaps several different disciplines, are required to address the full scope. The problem cannot be resolved in the usual three to six months needed for a Lean Six Sigma project. Hoerl and Snee (2017) gave the example of developing a default prediction methodology to protect a $500 billion portfolio at GE Capital. Several disciplines, including quantitative finance, statistics, operations research, and computer science, were needed to find an effective solution to this problem.
Complex and Challenging
The problem has significant complexity; it is not only technically difficult, but it has many facets (for example, political, legal, or organizational challenges in addition to technical challenges). Kandel et al. (2012) noted the dearth of research on how analytics are actually utilized within an organizational context and pointed out that this is critical to results. Typically, the technical problem cannot be addressed without understanding and addressing the nontechnical challenges. In the case of the default predictor, GE Capital insisted that, no matter how complex the predictor technology was, it had to provide clear and relatively simple advice to portfolio managers, without technical jargon.
Lack of Structure
The problem itself is not well defined, at least not initially. Many of GE’s original Six Sigma projects faced this issue, which led to the addition of the Define step in the DMAIC process. However, large, complex, unstructured problems often have an even greater lack of structure initially, requiring more up-front effort to structure the problem. In the case of the default prediction system, the word default does not have a generally accepted definition within financial circles. For example, Standard and Poor’s (S&P) uses a different definition of default than Dunn and Bradstreet (D&B). Therefore the team was asked to predict something that was not even defined.
Data Challenges
Most textbook problems in virtually all quantitative disciplines, from statistics to mechanical engineering, to econometrics, come with “canned” data sets. These might, of course, involve real data, but they are typically freely given with no effort required to obtain them. Generally, the data is of unquestionable quality, often presented in statistics texts as a “random sample.” Naturally, researchers and practitioners who have to collect their own data understand how challenging it is to obtain high-quality data. Random sampling is an ideal that is rarely accomplished in practice. For many significant problems, the existing data either is wholly inadequate or comes from disparate data sources of different quality and quantity. In the default prediction case study, minimal data existed because GE Capital was largely built by acquisition; there was no “master” set of data.
Lack of a Single “Correct” Solution
Most textbook problems also have a single correct answer, often given in the appendix. Many real problems also have a single correct answer. Even in online data competitions, such as those on kaggle.com, an objective metric, such as residual standard error when predicting a holdout data set, typically is used to define the “best” model. However, complex problems do not have a “correct” solution that we can look up in the appendix or even a reference text. They are too big, too complicated, and too constrained because of the issues discussed previously. Certainly, some solutions might be better than others, but the problem also could be solved in multiple ways; saying that one particular solution was “correct” or “best” would be virtually impossible.
The Need for a Strategy
Given the issues noted earlier, it is not possible to theoretically derive the correct solution for large, complex, and unstructured problems. Readers with a theoretical mind-set might be frustrated with the inability to find an optimal solution. Conversely, however, working solely by experience (trying to replicate previous solutions, for example) is generally hit-or-miss because each problem has specific complexities that make it different. A unique strategy needs to be developed to attack this specific problem, based on its unique circumstances. Both theory and experience help quite a bit, but an overall strategy that involves multiple tools applied in some logical sequence (and perhaps one that integrates multiple disciplines, especially computer science) is required.
In this sense, we argue that a solution needs to be engineered using known science, and often including statistics, as the components. In our opinion, textbooks and academic courses across quantitative disciplines do not discuss strategy enough, if at all. Engineering solutions to large, complex, unstructured problems requires a problem solving mind-set, but it must be guided by both theory and experience. A problem-solving mind-set typically looks for an effective solution, realizing that an optimal solution does not exist for such problems. As we discuss shortly, this typically requires an engineering paradigm versus a pure science paradigm.
Statistical engineering (Hoerl and Snee 2017) has been proposed as an overall approach to developing a strategy to attack such problems. Statistical engineering is defined as “the study of how to best utilize statistical concepts, methods and tools, and integrate them with information technology and other relevant disciplines, to achieve enhanced results” (Hoerl and Snee, 2010, p.12). Note that statistical engineering is not a problem solving methodology, per se, as Lean Six Sigma is, but rather is a discipline. However, a generic statistical engineering framework to attack large, complex, unstructured problems was given by DiBenedetto et al. (2014). Figure 1.1 shows this framework.
)
Figure 1.1 Framework for attacking large, complex, unstructured problems
In Figure 1.1, note that once the high-impact problem has been identified, it needs to be properly structured. Significant time and effort are typically required to understand the context of the problem. Large, complex problems have defied solution for a reason; a thorough understanding of the context of the problem is critical to finding a solution. By context, we mean such aspects as these:
The full scope of the problem, including technical, political, legal, and social aspects, to name just a few
How the problem came into existence in the first place
What solutions have been attempted previously
Why these solutions have been inadequate
Clearly, fixing a leak in an oil line does not require understanding this depth of context. However, addressing problems such as the Millennium Development Goals, discussed previously, certainly does. This is another illustration of why the solution needs to be tailored to the problem. Once the context is properly understood, the team is in a position to develop a strategy to attack this particular problem. The unique strategy typically entails several tactics or elements of the overall strategy. We discuss this framework in greater detail in Chapter 3, “Key Methodologies in a Holistic Improvement System.” See DiBenedetto et al. (2014) or Hoerl and Snee (2017) for further details on this framework.
Does Not Take Advantage of Big Data Analytics
Recent advances in information technology (IT) have led to a revolution in the capability to acquire, store, and process data. Data is being collected at an ever-increasing pace through social media, online transactions, and scientific research. According to IBM, 1.6 zetabytes (1021 bytes) of digital data are now available. That’s a lot of data—enough to watch high-definition TV for 47,000 years straight (Ebbers, 2013)! Hardware, software, and statistical technologies to process, store, and analyze this data deluge have also advanced, creating new opportunities for analytics.
At the beginning of the new millennium, the book Competing on Analytics (Davenport and Harris, 2007) foretold the potential impact data analytics might have in the business world. Shortly thereafter, Netflix announced a $1,000,000 prize for anyone who could develop a model to predict its movie ratings at least 10 percent better than its current model (Amartriain and Basilico, 2012). Picking up on the popularity of this challenge, the website kaggle.com emerged as a host to online data analysis competitions and became what might be called the “eBay of analytics.” Through kaggle, organizations that lack high-powered analytics teams can still benefit from sophisticated analytics by sponsoring data analysis competitions involving their data.
Further demonstrating the power of data and analytics, in 2011, the IBM computer Watson defeated human champions in the televised game show Jeopardy!. Data science has emerged and grown rapidly as a discipline to help address the technical challenges of Big Data (Hardin et al., 2015). In fact, Davenport and Patil (2012) have described data scientists as having the “sexiest job of the 21st century.” Programming languages such as R and Python have grown in popularity, and new methods have been developed to handle the massive data sets that are becoming more common. We’re referring here to computer-intensive methods such as neural networks, support vector machines, and random forests, to name just a few (James et al., 2013).
There is a “dark side” to Big Data analytics, however. As noted by Hoerl et al. (2014), the initial success and growth of Big Data led many to believe that combining large data sets and sophisticated analytics guarantees success. This naïve approach has proven false, with several highly publicized failures of Big Data. For example, Google developed a model to rapidly predict outbreaks of the flu based on people googling words related to the flu, such as flu, fever, sneezing, and so on (Lazar et al., 2014). This model, which appeared to detect flu outbreaks faster than hospitals detected them, was an early poster child for the power of Big Data analytics. However, Lazar et al. went on to point out that the predictive capabilities of the Google model have deteriorated significantly since its original development, to the point at which now a simple weighted moving average performs better.
Our point is not to disparage the potential of Big Data analytics, but rather to point out that coding and sophisticated analytics have not replaced the need for critical thinking and fundamentals. Studying coding and algorithm development is important and quite useful in practice. However, such study does not replace study of the problem solving process, statistical engineering, or continuous improvement principles. For example, as we said during our previous discussion of large, complex, unstructured problems, such problems have no optimal solutions. Therefore, we cannot develop an optimal algorithm to solve them; an overall, sequential approach involving several disciplines is typically required. To be sure, computer science is a key discipline that needs to be involved, but it is not the only needed discipline.
Cathy O’Neil, a self-described data scientist with a Ph.D. in mathematics from Harvard, wrote in more detail about the dark side of Big Data analytics in her uniquely named book Weapons of Math Destruction: How Big Data Increases Inequality and Threatens Democracy (O’Neil, 2016). Her fundamental point is not that data analytics or large data sets are inherently bad, but rather that programmers who are not properly trained in the fundamentals of modeling (including the limitations and caveats of models) are rapidly producing “opaque, unregulated, and uncontestable” models that are often blindly accepted as valid. Such invalid models are often applied to such important decisions as approving loans, granting parole, and evaluating employees or candidates.
Big Data and data science are frequently discussed in the context of analytics, statistics, machine learning, or computer science. In our experience, however, Big Data is rarely discussed in the context of continuous improvement. The reasons for this omission are not clear to us. Certainly, massive data sets provide unique opportunities to make improvements—assuming, of course, that they contain the right data to solve the problem at hand and that subsequent models go through appropriate vetting. Furthermore, newer, more sophisticated analytics, such as those mentioned earlier, provide additional options to consider when attacking problems, particularly the large, complex, and unstructured problems that cannot be easily solved with traditional methods. We feel strongly the Big Data analytics provide a significant opportunity for expanding both the scope and impact of continuous improvement initiatives.
Does Not Address Modern Risk Management Issues
As noted previously, the world certainly seems to be a more dangerous place than in the past, especially for business. Clearly, concerns over terrorism are not restricted to military or government institutions. For example, could Walt Disney have ever imagined the need for families coming to see Mickey Mouse to go through metal detectors and security checks? Yet they do, and for good reason. Financial institutions, energy companies, and businesses performing medical research on animals are frequent targets of threats of violence.
Of course, terrorism is not the only cause for concern from a risk management point of view. Identity theft is now a billion-dollar criminal enterprise in the United States alone. In many cases, the cost of illegal transactions, whether they are purchases with fake credit cards, fraudulent loans, or some other means, is borne not by the individual whose identity was stolen, but instead by the business that provided the loan or guaranteed the credit card purchase.
A more modern phenomenon is computer systems being hacked to obtain confidential information. As noted previously, the hack of Target’s credit card database not only allowed 40 million credit card numbers to be stolen, but also did irreparable harm to Target’s image. Beyond credit card numbers, organizations such as WikiLeaks (wikileaks.org) are more than eager to obtain and make public damaging information about businesses, including email exchanges, financial reports, and confidential legal documents. WikiLeaks generally obtains its material from other sources. Edward Snowden’s top-secret information (https://en.wikipedia.org/wiki/Edward_Snowden) concerning the National Security Agency (NSA) is perhaps the most obvious example.
Along the same lines, confidential emails among members of the Democratic National Committee (DNC) were hacked in 2016 and published on WikiLeaks. This caused significant embarrassment for the DNC, including the revelation of efforts by Chair Debbie Wasserman Schultz to favor Hillary Rodham Clinton’s campaign at the expense of Bernie Schultz’s. On July 25, 2016, Wasserman Schultz resigned her position because of the revelation on WikiLeaks. Considerable drama surrounded the original DNC hack, including accusations that the Russian government orchestrated the hack in an effort to enhance Donald Trump’s chances in the U.S. election (Sanger and Savage, 2016).
Clearly, businesses in the twenty-first century face unique security challenges, in addition to traditional business risks such as major lawsuits, environmental disasters, and catastrophic product failures. Therefore, risk management has become an even more critical business priority. What methods and approaches should be used for managing these types of risks, both the traditional ones and also the newer risks? We argue that the answer to this question is not obvious. However, it must be answered because the cost of failure in risk management is too high. Therefore, we consider risk management to be an integral element within a holistic improvement system.
