Process Variation—Enemy and Opportunity
by Ronald D. Snee
As the giants of scientific management and the quality movement long ago pointed out, work takes place in a series of interconnected processes.
Improving the way we work entails controlling and improving those processes, whether they are manufacturing ones such as pellet extrusion in pharmaceuticals, or business activities, such as loan application process in financial services.
However, it’s impossible to control and improve a process you do not understand.
Such understanding begins with variation—a venerable subject in statistical literature, but one that is sometimes misunderstood. Too often, variation is seen merely as the enemy that must be reduced whenever it appears. But unless we achieve a full and detailed understanding of variation, it inevitably reappears and worsens. Dealing with it becomes like the arcade game Whac-a-Mole—you hammer it whenever it appears, only to have it pop up in other places.
Instead of regarding variation as the enemy, it is more productive to see it as an opportunity. By fully identifying, characterizing, quantifying and reducing variation, you can improve the bottom line through reduced operating costs and improve customer satisfaction through better service.
You also can lift employee morale, a frequently overlooked benefit of variation reduction. People do not like to do rework, fail at their tasks or incur the ire of customers. Interestingly, employee satisfaction has been found to be a good predictor of customer satisfaction. Six Sigma can be effective approach for reducing variation and improving processes.1,2
It is critical to distinguish between improving the process average and reducing variation. As Jack Welch noted, “We have tended to use all our energy and Six Sigma science to move the mean. The problem is, as has been said, the mean never happens. The customer only feels the variance that we have not yet removed.”3
Consider two suppliers whose days to delivery
averages are almost identical but whose variations in days to
deliver differ (see Figure 1). Most customers would prefer the
consistency of supplier B to the wide variations of supplier A
even though supplier B takes a day longer, on average, to
deliver. In fact, if the customer is not ready to receive the
product, delivery too early can sometimes be worse than too
By focusing on variation reduction rather than the process average, supplier B is likely to enjoy higher customer satisfaction and in the long run win greater market share. As W. Edwards Deming put it, “If I had to reduce my message for management to just a few words, I’d say it all had to do with reducing variation.”4
Laws of Variation
Understanding variation should be a core competency of statisticians, Six Sigma Green and Black belts, quality engineers, quality managers and all other improvement professionals. Charged with improving processes, these professionals should think deeply about variation and its effects.
Statistical thinking is particularly helpful because its three key elements—process, variation and data—include the process that produces the variation, the sources of the variation and the use of data to deal with it.5,6 Further, as professionals approach improvement work, they also should be guided by the laws of variation:
- Variation is a fact of life—it is all around us and present in everything we do.
- All variation is caused.
- Variation can be predicted.
- Sources of variation are additive.
- Variation can be quantified.
- A small number of sources of variation contribute most of the variation.
- Process data contain variation produced by both the process and the measurement system.
- Process input variation affects process output variation.
- Variation affects managerial quality.
The process/variation perspective provides the context for problem solving and process improvement work as well as the pedigree of the data. Most importantly, the process view increases the likelihood problems will be solved successfully.
Separating Process Variation From Measurement Variation
There are two distinct kinds of processes of interest in improvement efforts. Processes that produce:
- The product or service provided by the organization to its customers.
- The data used to measure and monitor the performance of the process.
In seeking to understand the causes of variation, it is imperative to sharply distinguish between these two kinds of processes and consider both as possible sources of problems. Generally, statisticians and quality professionals need to pay more attention to the second kind of process—the one that generates the data.7,8
Consider a case in which process performance data variy widely. For example, the manufacturing department says the variation is due to the measurements coming from the analytical lab. The analytical lab says the process is too variable and out of control. Who’s right? Fortunately, tools exist to separate process variation from measurement variation. For example, if studies show the gage repeatability and reproducibility (R&R) variation is less than 30% of the total variation, the measurement system is acceptable for guiding process improvement work.
In the case of destructive tests, in which the sample is destroyed in running the test, nested sampling designs are used to assess the validity of the measurement system.9 In one study, a measurement system variation represented 81% of the total variation.10 The root cause was found to be the failure to follow proper sampling procedures.
In a subsequent study in which the sampling procedures were followed, the measurement system variation was reduced by 41%. Here the sampling procedure is considered part of the testing method.
It should be remembered, too, that variation affects process flow. The defects and rework caused by variation result in increased inventory, more movement of material, overproduction, wasted motion and wasted production time—all significant sources of slower process flow as well as reduced customer satisfaction and increased costs. Sensitivity to the impact of variation on process flow should therefore be part of lean studies of work processes.
How to Look For Sources of Variation
Several approaches can be used to find key sources of process variation including:
- Mapping the process, identifying the key input and output variables at each step in the process, and using a cause and effect matrix to prioritize the input variables to find the most important ones.
- Mapping the process, brainstorming a list of key input variables and using a fishbone diagram to identify the most important of them. This approach works best when there is only one process output variable of interest.
- Making a control chart of key process output variables and looking for out of control signals indicating special cause variation, which will identify key sources of variation.
- Conducting a multi-vari study in which data are collected on the process as it operates.11 Analysis of the data will detect key sources of variation due to process variables and uncontrolled noise variables such as sources of process input material, operating teams and different pieces of equipment. Process mapping and identifying process variables via brainstorming or using the cause and effect matrix are useful approaches to identifying the variables to be included in a multi-vari study.
Where to Look for Variation
In looking for sources of variation there are some usual suspects, most notably material lots, machines, operating lines and human behavior. The guiding rule is all potential sources of variation are guilty until the data prove them innocent.
For example, a biopharmaceutical process was producing lower than expected yields, causing concern the process would be unable to meet market demand. A control chart analysis of the yields of the previous 57 batches showed a big difference in the yields of the different raw material batches used in the process. The specifications on the raw material batches were tightened, and the process yield increased by 25%, significantly adding to the bottom line as well as enabling the process to meet customer demand.
Machines are often a potent source of variations. Ellis R. Ott, quality pioneer and ASQ honorary member, pointed out on many occasions that, “Where there are three or four machines one will be substantially better or worse than the others.” Sim-ilarly, different operating lines—whether they are production lines in manufacturing or a standardized business process—that are apparently identical in every respect nevertheless might produce variation.
Often variation in operating lines—both in manufacturing and services—results from variations in human behavior. Different people have different methods and styles of working, learning and thinking. They take in, process and communicate information in different ways. Moreover, people differ from day to day, group to group and organization to organization.
The effect of the “Monday morning blues” on the work of individuals on Monday vs. the rest of the week has been well documented. People might differ from group to group as a result of cultural, educational and occupational backgrounds that affect their work. People in a manufacturing organization might pay far more attention to manufacturability while people in product design might care more about style or product features. The list of possible differences is nearly endless.
For example, a manufacturing company had undertaken studies to help increase the output of a process. Analyses of the data showed none of the variables studied had any significant effect, yet there was considerable variation in the output of the process. Discussion then turned to the four teams of operators that ran the process.
Analysis of the effects of these different teams revealed a major portion of the variation in output was due to one team’s having production levels considerably higher than the other three, which had essentially the same output level. Not surprisingly, the team with the highest output also was the most experienced.
One of the most effective ways to reduce variation is to anticipate it and then prevent it by designing processes, products and management practices to be insensitive to uncontrollable variation. This concept of insensitivity to variation, often called robustness, is a key aspect of statistical thinking, yet it is greatly underused.12
As the quality movement has taught us, 100% quality cannot be inspected in after the fact; it should be designed in from the inception of a product, process or management practice. A robust product can be designed to be insensitive to variations in conditions of manufacture, distribution, use and disposal. A robust process should be insensitive to uncontrollable variations in process inputs, external factors and transformations in the course of the process such as activities, steps, process variables and uncontrolled noise variables.
To create robust management practices, you should develop strategies relatively insensitive to economic trends and cycles. You also should design a project system insensitive to changes in personnel and project scope. You also can create working conditions that take into account differing employee needs, such as flexible work hours, and you can enable personnel to adapt to changing business needs.
How to Know When You Understand a Process
Whether you are anticipating variation in a future process or confronting it in an existing process, how do you know when you have achieved the full understanding of the process essential for controlling that variation? Simply put, you understand a process when it is possible to predict its future performance. A process is understood when:
- Critical variables (X’s) that drive the process are known.
- Critical uncontrolled (noise) variables that affect the process output are known, and the process has been designed to be insensitive to these uncontrolled variations.
- Measurement systems for process variables (X’s) and outputs (Y’s) are in place and the amount of measurement variation is known.
- Process capability is known.
- Effective process control procedures and control plans are in place.
With the understanding of process variation as the foundation of process understanding, improvement professionals then can be continually on the lookout for important sources of variation—whether in the process itself or the measurement process—and confidently reduce that variation, secure in the knowledge they can significantly improve performance.
- R.D. Snee and R.W. Hoerl, Leading Six Sigma: A Step-by-Step Guide Based on Experience With GE and Other Six Sigma Companies, Financial Times Prentice Hall, 2003.
- R.D. Snee and R.W. Hoerl, Six Sigma Beyond the Factory Floor: Deployment Strategies for Fin-ancial Services, Health Care and the Rest of the Real Economy, Financial Times Prentice Hall, 2005.
- General Electric Co., 1998 Annual Report, www.ge.com/annual98/pdf/full.pdf.
- Henry R. Neave, The Deming Dimension, SPC Press, 1990.
- ASQ Statistical Division, Glossary and Tables for Statistical Quality Control, third edition, ASQ Quality Press, 1996.
- R.W. Hoerl and R.D. Snee, Statistical Thinking: Improving Business Performance, Duxbury Press, 2002.
- R.D. Snee, “Are We Making Decisions in a Fog? The Measurement Process Must Be Continually Measured, Monitored and Im-proved,” Quality Progress, Vol. 38, No. 12, 2005, pp. 75-77.
- R.D. Snee, “If You’re Not Keeping Score, It’s Just Practice. If You’re Not Measuring, You’re Not Competing,” Quality Progress, Vol. 39, No. 5, 2006, pp. 72-74.
- R.D. Snee, “Graphical Analysis of Process Variation Studies.” Journal of Quality Technology, Vol. 15, 1983, pp. 76-88.
- G.E.P. Box, J.S. Hunter and W.G. Hunter, Statistics for Experimenters: Design, Innovation and Discovery, second edition, Wiley-Interscience, 2005.
- R.D. Snee, “My Process is Too Variable, Now What Do I Do?” Quality Progress, Vol. 34, No. 12, 2001, pp. 65-68.
- R.D. Snee, “Creating Robust Work Processes,” Quality Progress, Vol. 26, No. 2, 1993, pp. 37-41.
RONALD D. SNEE is principal of performance excellence and lean Six Sigma initiative leader at Tunnell Consulting in King of Prussia, PA. He has a doctorate in applied and mathematical statistics from Rutgers University in New Brunswick, NJ. Snee has received the ASQ Shewhart and Grant medals and is an ASQ fellow.