"Operational Excellence" (OE) is a term now in common usage in the business world. But what exactly is OE? Before defining OE, we must put it into context. OE involves efforts at the level of individual processes within an organization. In practice, it demands that any process be designed, executed and improved according to Deming's Plan/Do/Check/Act (PDCA) loop  (Figure 1).
In this framework, the "Plan" phase contains the process design and how it should be executed to achieve optimal performance, which often is called the control plan or control strategy. Once the plan is in place, the process is run (the "Do" phase). In an ideal world, the plan would be perfect and its execution would be flawless — the process would deliver optimal performance from day one. However, in reality, this never is true and process performance isn't optimal. So, we use the "Check" phase to compare actual with desired performance, and decide what actions must be taken to close the performance gap. These actions then are implemented in the "Act" phase.
OE generally is thought of as applying to the Do/Check/Act phases. However, because the Plan phase is the output of other processes that have their own PDCA loops, in a general way OE applies to all phases of the loop. So, OE can be viewed from two perspectives — an end in mind (the goal of optimal process performance) and a way to achieve it (flawless execution and, when issues and opportunities occur, the ability to recognize them and make improvements).
Given the above, the following definition seems reasonable: For any organization or part of that organization, OE is being able to reliably produce great quality at low cost while complying with all regulatory requirements.
This definition focuses on all the results of the process. It doesn't involve choosing one desired performance measure over another. Rather, it requires that all goals be met to truly achieve OE. The term "quality" reflects the degree to which the output of the process satisfies internal and external customer expectations. The word "reliably" is perhaps the most important word in the definition simply because many times it's the part of OE that's not recognized. However, performance reliability is a hallmark of OE. It's not only about optimal results but also about achieving them consistently over time. Such reliability is fundamental to a good customer/supplier relationship. Indeed, it's why process stability (statistical control) must be established in a process before process capability can be quantified [2,3].
OE represents an ideal that never can be fully reached in practice. Nevertheless, an organization should strive for OE. As legendary football coach Vince Lombardi said, "Perfection is not attainable. But if we chase perfection, we can catch excellence." Customer expectations always are changing, as are those of society (as reflected in evolving regulatory requirements), so at best any performance goal is temporary. Given the time it takes for an organization to make significant improvement, by the time the change has occurred the goal already may be out of date. As an example, in the past some companies considered three-sigma performance as world class but now strive for six sigma — and, for high-consequence situations, even six sigma may not suffice.
Given that goals are changing and reliability is so important, the second view of OE becomes critical. That is to say, as well as OE representing an end in mind, its never-attainable nature begs for an investigation of "how" it is attained. Having a solid approach to achieving the goals of OE means that reliability can be built into a process and the organization can respond quicker to changing requirements. However, actual performance is the only way to judge the success of efforts. If an organization is using what it considers to be best-in-class approaches to improve OE but defect rates still are very high and not improving, then clearly its approaches are flawed and must change.
People execute and manage processes — so, the core of the approach to OE must focus on their behaviors. Such behaviors are the easiest way to see OE when it's being done right (and when it's being done wrong). In addition, we must recognize that behaviors don't just happen but are caused. Therefore, we also must look at the elements that must be in place to promote the right behaviors to support OE and minimize the risk of the wrong behaviors that undermine OE.
One way to think about this is to consider how organizations react to human error issues. One approach is simply to consider such errors to be root causes and punish the individuals involved. A more-enlightened approach is to ask why such errors are occurring and whether organizational elements that support people — e.g., training, timeliness of important process data, suppression of nuisance alarms, etc. — are adequate. This view treats improper behaviors as symptoms of gaps in the underlying systems.
So, what elements must be in place for OE to be a success? Fortunately, there's an organizing principle that describes the required organizational elements in a useful way. It's called the 7S model.
THE 7S MODEL
This model (Figure 2) was created in 1980 to show the elements that must be in place for positive organizational change to occur . It defines all elements that must be part of the change and emphasizes that no element on its own is enough.
If an organization strives to improve OE but fails to pay attention to all seven elements, it will not succeed. We'll focus on four of the elements — Structure, Strategy, Skills and Systems. We'll look at each as it applies to improving OE across the Do, Check and Act phases, along with some behaviors that may arise if the element isn't addressed.
Structure. Effectively implementing OE requires establishing an operations team whose main focus is to run a particular process according to the procedures identified in the Plan phase. Several such process teams report to a higher-level team (e.g., an area management one); all area teams in turn report to a site management lead team (Figure 3). An important aspect of this structure is the relationship between the different levels. Basically, each area team provides performance objectives and resources to its process teams; the process teams then report performance to the area team and request resources from it. (The process teams must have access to functional expertise — quality, safety, engineering and so on — because it would be unreasonable to expect them to have all this expertise.) The area team decides how to allocate resources among process teams based upon needs and opportunities. The same basic relationship exists between the area teams and site lead team.
Without a good structure in place to support implementation of OE, prioritization becomes more difficult and process teams may not be sure what are the most important tasks for them to work on. And those priorities may change quickly from day to day as new issues arise. I remember being at a process team meeting once where someone from outside the team was updating it on an issue needing attention. After the person gave the update and left the meeting, the team leader discussed the issue with the team and concluded the team still needed to focus first on the current priority list. In effect, the team leader was providing a firewall if someone later asked why the team wasn't giving this new issue its full and immediate attention.
Strategy. A good strategy for OE is to reduce variability in all aspects of our processes and use the reduced variation to accelerate continuous improvement [5,6]. A relentless pursuit of variability reduction is the foundation of the revolution that Deming brought to Japan . Highly variable processes are harder to improve because the variability can hide relationships; it takes longer to see the impact of changes and anticipating their potential unintended consequences is more difficult.
Probably the most fundamental behavior seen when OE implementation is low can be expressed by "if it's within specifications, we're okay." This means that variability reduction only is seen as an issue when an output is out of specification. This attitude also arises on the input side where an investigation is being performed after some processing issue. If the inputs, such as raw materials, meet their specifications (as attested by Certificates of Acceptance, for example), then it's assumed they aren't part of the problem.
Skills. People need certain skills to implement OE. These fall into three types:
1. Skills related to running the process — correctly understanding the work instructions, how to run equipment and so on;
2. Skills related to how the process works — adequately comprehending why the work instructions are written the way they are, how the equipment operates (the first principles of the process) and so on; and
3. Skills related to process monitoring — properly interpreting the data being produced by the process and responding appropriately.
All three types of skills are essential. For example, without the first set of skills, people may interpret the instructions differently, leading to higher variability in the process. Without the second set, people simply assume the current situation is ideal and so may miss opportunities for improving the process. Without the third, people may treat random variation — which occurs in all real processes — as a true change, unnecessarily tampering with a process and actually increasing its variation. (By the way, Deming said success depends upon creating generations of statistically minded engineers and scientists .)
Systems. Here, we mean the systems that support the process team as it runs the process. Systems can help in at least three ways:
1. Increasing efficiency and, hence, easing implementation of OE. For example, computer-based systems can update process data very frequently and store the data for long periods of time — a task not feasible with people-based systems.
2. Improving consistency. Embedding key decision algorithms in a system can eliminate variability caused by people. For instance, a system to create control charts for the process can ensure the correct control chart with correctly calculated control limits is used for each type of data being monitored.
3. Providing more-sophisticated analysis of the process. There are many different ways to analyze process data. Current developments in "machine learning" (where we train a system to understand what "normal behavior" looks like in a process so it can detect anomalies) are providing very sophisticated approaches that only can be implemented by a computer-based system [8,9].
Processes not supported by appropriate systems will exhibit higher variability and less efficiency. So, don't sacrifice OE objectives just to fit within the constraints of an existing system. Instead, design the appropriate system based upon the OE objectives.
Many years ago, we had a process that wasn't delivering the throughput required. This generated a lot of emphasis on the process. Many people were involved and were working really hard to improve the throughput — but with limited success. Then, a new manager for the line started and changed how the process team worked. This led to a significant improvement in the throughput.
He enhanced the level of OE in the process and the team in several ways:
• He changed the structure within the team so resources were more focused on certain tasks. The first part of the team concentrated on the day-to-day running of the process and dealing with minor issues that could be fixed quickly and easily. The second part zeroed in on solving the major special-cause variation problems while the third part tackled the major common-cause variation problems. The manager did his best to ensure that people kept focus on a problem until it had been resolved, not jumping from one issue to the next.
• He instituted a program to improve people's understanding of process variation so they were better able to differentiate between special- and common-cause variations. As part of this, he hired an external consultant to provide objective feedback on what they were doing and how they could do better.
• The process had a very good system for collecting data. However, the system's capabilities weren't being exploited. So, he spent time defining the ways in which people should look at the process data and how to present these data to the process team in ways that facilitated easy interpretation. This helped, for example, to better distinguish between special- and common-cause variations.
• He questioned the existing mental models about the impact of input variation on the variability of the process. For example, because there was a good set of work instructions, it had been assumed that people had minimal impact on variation. However, control charting the throughput data as a function of shifts showed that all the shifts were performing differently. This spurred a renewed focus on having all operators perform all tasks consistently — and had the added benefit of improving general communications between shifts, which is very important when it comes to reducing process variability. It also had been assumed that the variation of the input raw materials was insignificant. However, he was able to demonstrate this wasn't the case for one key input raw material; modifying the control strategy around this raw material led to improvement.
OE is an objective that can be defined in terms of expectations for high levels of process performance. However, it also must encompass the behaviors that support and sustain such enhanced performance over long periods of time, engraining them into the DNA of the organization. These behaviors only will occur if certain key elements (as outlined by the 7S model) are improved as part of the journey towards higher levels of OE.
BERNARD MCGARVEY is a senior engineering advisor for Eli Lilly and Company, Indianapolis, Ind. E-mail him at firstname.lastname@example.org.
1. Deming, W. E., "Out of the Crisis," M.I.T. Press, Cambridge, Mass. (2000).
2. McConnell, J. S., "Analysis and Control of Variation", 4th ed., American Overseas Book Co., Norwood, N.J. (1987)
3. Wheeler, D. J. and Chambers, D. S., "Understanding Statistical Process Control," 2nd ed., SPC Press, Knoxville, Tenn. (1992).
4. Waterman, R. H., Peters, T. J. and Philips, J. R., "Structure is not Organization," Business Horizons, Vol. 23, No. 3, pp. 14–26 (1980).
5. Wheeler, D. J., "Advanced Topics in Statistical Process Control," SPC Press, Knoxville, Tenn. (1995).
6. McConnell, J. S., Nunnally, B. and McGarvey, B., "Meeting Specifications Is Not Good Enough — The Taguchi Loss Function," J. Validation Technology, Spring 2011.
7. Deming, W.E., "On Some Statistical Aids Toward Economic Production," Interfaces, Vol. 5, No. 4, pp. 1–15 (1975).
8. Mitchell, T. M., "Machine Learning," McGraw-Hill, New York City (1997).
9. Shmueli, G., Patel, N. R. and Bruce P. C., "Data Mining for Business Intelligence," 2nd ed., John Wiley & Sons, Hoboken, N.J. (2010).