Six Sigma and Project Management
Six Sigma is just making its appearance in project management. Six Sigma is the name coined by Motorola in the 1980s for a process and throughput improvement strategy it developed from some of the process control work done originally in the Bell Laboratories in the 1920s and later taken to Japan in the 1950s by W. Edwards Deming. Six Sigma's goal is to reduce the product errors experienced by customers and to improve the quality of products as seen and used by customers. Employing Six Sigma throughout Motorola led, in part, to its winning the prestigious Malcolm Baldrige National Quality Award 1988.
The name "Six Sigma" has an origin in statistics. The word "sigma" we recognize as the Greek letter "s" which we know as a. From our study of statistics, we know that the standard deviation of a probability distribution is denoted by σ. Furthermore, we know that the confidence measures associated with a Normal distribution are usually cast in terms of so many σ from the mean: for example, as said many times before in this book, ±1σ from the mean includes about 68.3% of the possible values of a random variable with a Normal distribution. Most project managers are comfortable with "2σ" estimating, which covers over 95% of all outcomes, and some project managers refine matters to "3σ", which encompasses a little over 99.7% of all outcomes. Six Sigma seems to be about going ±"6σ" from the mean and looking at the confidence that virtually all outcomes are accounted for in the analysis. In reality only ±4.5σ is required and practiced in the Six Sigma program, as we will see.
Six Sigma and Process Capability
Six Sigma is a process capability (Cp) technique. What is meant by process capability? Every man-made process has some inherent errors that are irreducible. Other errors creep in over time as the process is repeated many times, such as the error that might be introduced by tool wear as the tool is used many times in production. The inherent errors and the allowable error "creep" are captured in what is called the "engineering tolerance" of the process. Staying within the engineering tolerances is what is expected of a capable process. If the Normal distribution is laid on a capable process, as shown in Figure 8-8, in such manner that the confidence limits of the Normal distribution conform to the expectations of the process, then process engineers say with confidence that the process will perform to the required specification.
Figure 8-8: Capable Process and Normal Distribution.
A refinement of the process capability was to observe and accommodate a bias in position of the process mean. In other words, in addition to small natural random effects that are irreducible, and the additional process errors that are within the engineering tolerance, there is also the possibility that the mean of the process will drift over time, creating a bias. A capable process with such a characteristic is denoted Cpk.
3.4 Parts Per Million
In the Motorola process, as practiced, the process mean is allowed to drift up to 1.5σ in either direction, and the process random effects should stay within an additional ±3σ from the mean at all times. Thus, in the limit, the engineering tolerance must allow for a total of ±4.5σ from the mean. At ±4.5σ, the confidence is so high that speaking in percentages, as we have done to this point, is very awkward. Therefore, one interesting contribution made by the promoters of Six Sigma was to move the conversation of confidence away from the idea of percentages and toward the idea of "errors per million opportunities for error." At the limit of ±4.5σ, the confidence level in traditional form is 99.9993198% or, as often said in engineering shorthand, "five nines."
However, in the Six Sigma parlance, the process engineers recognize that the tails of the Normal distribution, beyond 4.5σ in both directions, hold together only 6.8 * 10-6 of the area under the Normal curve:
Total area under the Normal curve = 1
1 - 0.999993198 = 0.00000680161 = 6.8 * 10-6
Each tail, being symmetrical on each side, holds only 3.4 * 10-6 of the area as shown in Figure 8-9.  Thus, the mantra of the Six Sigma program was set at having the engineering tolerance encompass all outcomes except those beyond 4.5σ of the mean. In effect, the confidence that no outcome will occur in the forbidden bands is such that only 3.4 out-of-tolerance outcomes (errors) will occur in either direction for every 1 million opportunities. The statement is usually shortened to "plus or minus 3.4 parts per million," with the word "part" referring to something countable, including an opportunity, and the dimension that goes with the word "million" is silently implied but of course the dimension is "parts."
Figure 8-9: 3.4 Parts Per Million.
The move from 4.5σ to Six Sigma is more marketing and promotion than anything else. The standard remains 3.4 parts per million.
Six Sigma Processes
In one sense, Six Sigma fits very well with project management because it is a repeatable methodology and the statistics described above are the outcome of a multistep process not unlike many in project management. Summarizing Six Sigma at a high level:
Define the problem as observed in the business or by customers and suppliers
Define the process or processes that are touched by the problem
List possible causes and effects that lead toward or cause the problem within the process
Collect data according to an experiment designed and developed to work against the causes and effects in the possible cause list
Analyze what is collected
Exploit what is discovered in analysis by designing and implementing solutions to the identified problem
Project Management and Six Sigma
Looking at the project management body of knowledge, especially as set forth by the Project Management Institute® virtually every process area of project management has a multistep approach that is similar in concept to the high-level Six Sigma steps just described. Generally, project management is about defining the scope and requirements, developing possible approaches to implementing the scope, estimating the causes and effects of performance, performing, measuring the performance, and then exploiting all efforts by delivering product and services to the project sponsor.
The differences arise from the fact that a project is a one-time endeavor never to be exactly repeated, and Six Sigma is a strategy for repeated processes. Moreover, project managers do a lot of work and make a lot of progress at reasonable cost with engineering-quality estimates of few parts per hundred (2σ) or perhaps a few parts per thousand (3σ). However, even though projects are really only done once, project managers routinely use simulation to run projects virtually hundreds and thousands of times. Thus, there is some data collection and analysis at the level of few parts per thousand though there may be many millions of data elements in the simulation.
The WBS and schedule network on very large projects can run to many thousands of work packages, perhaps even tens of thousands of work packages, and thousands of network tasks, but rare is the project that would generate meaningful data to the level of 4.5σ.
The main contribution to projects is not in the transference to projects of process control techniques applicable to highly repetitive processes, but rather the mind-set of high quality being self-paying and having immeasurable good consequences down the road. In this sense, quality is broadly dimensioned, encompassing the ideas of timeliness, functional fit, environmental fit, no scrap, no rework, and no nonvalue-added work.
Six Sigma stresses the idea of high-quality repeatability that plays well with the emerging maturity model standards for project management. Maturity models in software engineering have been around for more than a generation, first promoted by the Software Engineering Institute® in the early 1980s as means to improve the state of software engineering and obtain more predictable results at a more predictable resource consumption than heretofore was possible. In this same time frame, various estimating models came along that depended on past projects for historical data of performance so that future performance could be forecast. A forecast model coupled with a repeatable process was a powerful stimulus to the software industry to improve its performance.
In like manner, the maturity model and the concepts of quality grounded in statistical methods will prove simulative to the project management community and will likely result in more effective projects.
A notation common with very small numbers, 3.4 * 10-6 means "3.4 times 1/1,000,000" or "3.4 times one millionth." In many other venues, including computer spreadsheets, the letter E is used in place of the "10" and the notation would be "3.4 E-6." Another way to refer to small numbers like this is to say "3.4 parts per million parts," usually shortened to "3.4 parts per million." A "part" is anything countable, and a part can be simply an opportunity for an event rather than a tangible gadget.