11.4 Activities of the Control Phase

Up to now, you have learned what it takes to define, measure, analyze, and improve. In other words, you have, for the most part, determined just what is the necessary course of action to take to ensure that you can execute. You have also put your plan in motion. “Control” is all about how to make sure that your plan is working as intended even after you hand it off to the process owner.

When we say “control,” what we don’t mean is everything that happens must be at an exact precision. For instance, basketball players don’t aim for “nothing but net” with every shot. Sometimes the ball bounces on the rim and rolls in. Other times the ball banks on the backboard before falling in the basket. Players with great field goal percentages tend to have an innate feel for whether their shot will likely go in based on a combination of the height of their release, how hard they flick their wrist, their aim, their follow-through. Yet, all of these practices cannot fully translate into in-game performance without also studying game films to understand opponent tendencies. In other words, getting down the motions alone cannot allow a process to stay under control. You have to know the external factors as well! Collectively, this combination of input and output factors form the basis of the conclusion of your team’s DMAIC journey:

1. What are the risk factors that can threaten a process to be out of control?

2. What are the key process input variable (KPIV) and key process output variable (KPOV) controls?

3. How do you monitor the process?

Understanding these factors, along with knowing your internal capabilities both before and after your project would allow you to come up with not only a monitoring plan but also a response plan! Ideally, at the end of the Control stage, you and your team would be ready to hand off the newly improved process to the process owner. Think of this stage as basically writing an instruction manual to tell whoever manages the process how everything works, what the vital parts of the process to monitor are, and what to do in the event the process shows signs of breaking down. At the end of this stage, all of the following needs to be done.

Next, let’s see just how we can put them together.

Monitor and Control Plan

Now, you want to make sure that the newly-established process can be sustained even after your project ends. Of course, you can’t just tell the next person “just keep things going the way they are.” In other words, you need to have a plan. Knowing what can go wrong is especially important for coming up with a monitor and control plan. For instance, a store manager needs to use the process map previously defined to see just what it takes to provide great customer service: products must be in stock; products must not be damaged prior to being displayed; customers must be able to find products in a logical location; customers must be served quickly at the check-out counter; prices must be accurate; cashiers need to exchange pleasantries with customers. Through proactive monitoring, you can systematically ensure that all the work you have done thus far will not be quickly and easily reverted. In fact, having a control plan allows you to monitor the most vital aspects of your process—like the cadence of your pedaling on a bicycle—in order to allow you to know when something is not going right. For instance, you may become distracted by the beautiful nature around you on your bike ride to work and begin to peddle slower, which may cause you to be late!

Of course, a control plan should not be put together haphazardly. A control plan is purposeful, useful, clear, and can be used to promote accountability ranging from suppliers to assembly line workers and their supervisors. Most importantly, you need to recognize just what it is you need to control. Are you trying to address an aspect of quality from the next-customer perspective or the end-user perspective? This question would be easy to answer if you make products directly sold to consumers, who are both your next-customer and end-user. Understanding this viewpoint can help you determine what aspect of quality to monitor and control. The typical control plan has the following four common elements:

1. Identification fields: First you need to know what it is you are trying to control and who are the responsible parties. If you are trying to address input quality, then your identification fields would focus on individual parts, the supplier of these parts, and information that link these parts to its purchase orders. If you are trying to address service quality, then these fields would need to identify the specific services as well as the parties responsible for carrying out these services.

2. Process information and equipment used: Next, you need to include information that would allow you to match the specific part or service that is the target of your control plan with your process flow diagram from earlier parts of your DMAIC cycle. This information allows you to use all quality management tools in conjunction.

3. Relevant characteristics of process output: This portion of your control plan must be catered to the specific process. In the example mentioned above involving retail management, would you proactively monitor the out-of-stock occurrences? Or would you rather skip directly to customer satisfaction? While every process has a wide variety of outputs, it is usually not feasible to monitor all of them. That is why Process Failure Mode Effect Analysis can be especially helpful in determining what you need to prioritize. More importantly, revisiting your FMEA helps you to illustrate to the new process owner both what your team managed to improve and what must be prioritized under the newly improved process.

4. Specifications and control method: Different process output requires different control method. You will need to decide the following: what are the specification limits under your current process for different KPIV and KPOV? Based on the nature of your specification limits, an appropriate control method could be designed.

5. Measurement: How can you measure different functional elements of quality control? Is it feasible to measure every single input and output? Each functional element might require customized ways to measure. Some might even require special tools to measure.

6. Reaction Plan: Although never intended, sometimes you do find either input or output variables to be outside the specification limits you defined for the process. What will you do? It depends on whether the defect is identified for an input or an output variable. If it is an input, then your option would most likely be return to the source. Often, this source is a supplier or vendor. Other times, the input could be the output of a previous step, in which case you just uncovered a previously undetected defect! On the other hand, if the defect is identified for an output variable, then you would want to alert the manager who is in charge in order to trigger a process review.

As you can see in the sample process control plan, the objective of a control plan is to define what components or processes you want to monitor, how to measures their specifications, what the specification limits are, how to sample, and finally, what to do in the event of an exception. You probably also noticed that “how to sample” usually involves measuring or evaluating a set quantity within a given allotment. For instance, the LCD screen in the sample is measured through visual inspection of 10 units per lot, whereas screws and casing are measured using sensors at 100 and 50 units per production shift. There’s a very good reason why we don’t inspect or assess every single component. That is because, unlike a basketball player who takes about 10 to 20 shot attempts per game, the typical retailer serves thousands of customers each day, and the typical manufacturer uses tens if not hundreds of thousands of components per shift: there simply isn’t enough time and resources for a company to measure every KPIV and KPOV.

Production Scale and Control

As we discussed in the control plan, businesses must assess and evaluate samples of their components rather than every single one of them. For instance, Apple sold almost 47 million iPhones in the world in the 4th quarter of 2018 alone! Surely, its assembly supplier, Foxconn, did not use its control plan to manually verify the production quality of every single one of the 47 million iPhones.

This is where random sampling and inferential statistics come in. Through both of these statistical tools, companies can continuously monitor their product and service quality in order to proactively identify and address problems residing in their production processes before they balloon to become unmanageable.

Thankfully, by this time, you have already done some of the work necessary to overcome obstacles presented by large-scale productions. That is because statistical sampling methods allow you to test hypotheses and draw conclusions regarding both your process improvements and your ability to maintain them. In general, your process outcomes can be grouped into two broad categories of measurements:

1. Attributes: These types of data are all about whether something possesses a certain characteristic. They tend to be measured in terms of parts per million (PPM) and defects per million opportunities (DPMO).

2. Variables: These types of data are characteristics that can take on a range of values. For instance, PepsiCo monitors the amount of soda filled into a bottle of soft drink. They tend to be measured based on normal distribution, such as the capability index (Cpk).

Do these sound familiar? They should. That is because you have performed the same calculations during the Measure phase of the DMAIC cycle. Whereas in the Measurement phase you learned how to apply these metrics to a production or service process, they are now being used once again in the Control phase for two primary purposes: For one, the metrics help you determine what is the current state of your newly-upgraded process. More importantly, establishing baselines for the superior process basically serves as the lanes on a road: if your process begins to drift across lane divider, the Control system would alert you to get back in lane to avoid collision!

Mistake Proofing and Poka-Yoke

Ever heard of the saying “What can go wrong, will go wrong?” Although the phrase is a little bit too general, it illustrates one thing: errors in any process, whether or not easily detected, will eventually result in faulty outcomes. The trick, then, is to prevent errors from occurring altogether. This is where mistake-proofing comes in. Mistake-proofing is the systematic use of devices or methods to eliminate or detect causes of an error either before or immediately after the error occurs. In other words, mistake-proofing is all about the pursuit of zero-defect.

Poka-yoke, a close synonym, is a term popularized by Shingeo Shingo within the Toyota Production System (TPS), and it has a narrower aim to eliminate unintentional human errors. What both mistake-proofing and poka-yoke have in common is that their aim is to first, prevent errors that result in defective production and service outcomes; second, to ensure there is a mechanism in place to detect errors when they do occur, since not all errors can be fully prevented; and third, to minimize the severity of a defect, since perfect quality is not attainable. Here is how they can be done:

1. Understand the process. Your first order of business is to know where errors might occur. Process mapping and flow charts you created in previous DMAIC phases are particularly handy for this.

2. Identify the cause. Applying the process failure mode and effects analysis (PFMEA) here can help you to understand what is most likely to be causing errors and help you prioritize your efforts.

3. Address the cause. Once you identified the cause, you need to consider what to do. Can you implement a corrective action to eliminate the error? If not, can you replace or eliminate that step in the process?

4. Ongoing detection. Although we strive to eliminate all errors, the reality is that perfection tends to be not attainable. Thus, you need to make sure that a mechanism is in place to be able to detect errors once they take place.

5. Implementation and Follow-up. In this step, you begin to test the solutions you identified in steps 3 and 4. If your pilot test appears successful, mass implementation may then be rolled out.

The above 5-step process should be understood as a cycle. That is because even the best process still results in defects. Hence, mistake-proofing should be ongoing to detect all errors and address their causes. In general, errors can be detected three different ways:

1. At the Source. This is a preventative measure and the most ideal point to set up an inspection point. By catching errors before defects occur, you are also saving lost time and costs.

2. During the Task. Secondary to preventing defects from occurring, the next best thing is to detect defects as they are taking place. This way, immediate feedback is transmitted to the source of the error to more quickly address them at the source.

3. After the Task. This is a purely reactive form of detection, which has a longer feedback loop to the source of the error.