In this article we will examine the ten mistakes companies make when attempting to launch / apply SPC. These are a distillation of my experience across multiple industries and dozens of companies I have worked with.
| Statistical Process Control is one of the tools that can be used by both Manufacturing, and Service organizations to deliver better quality to the customer. However, it is often misunderstood and misapplied. Mistake No. 1: It does not apply to us: we are different what we do is different the nature of our work is different therefore, SPC cannot be used here. You will come across cases where management, and / or others in the company feel that their operation is too unique, or different, and therefore SPC is not applicable. Such beliefs may arise because of several reasons such as: frequent process changes product changes short runs lack of metrics hesitation in setting up measurements / metrics about processes in the organization ignorance of the how it can work for the organization and the benefits it can deliver fear of how it might workout … If a process creates an output, SPC can be applied. Philip Crosby, a noted Quality Guru said, “All work is a process”. If there is an output, we will need to find a way to measure it, otherwise it becomes difficult to gage how well the output serves the customer’s needs. When we start measuring, data are collected – which can become the raw material for SPC. Mistake No. 1 – What You Can Do About It… Start with measuring outputs; most organizations find this to be the easiest place to start. Track the variation in outputs and see how it behaves. Concurrently work with the customer to either obtain or develop requirements / specifications. Then compare the variation in your outputs to customer specifications. Let the data tell you if your process is stable (operating within expected control limits), and capable (well within the specification limits). Keep in mind, to become more effective – eventually you will need to begin tracking the leading indicators. Staying put with measurement of outputs (lagging indicators) is likely to keep you in the reactionary (firefighting) mode, rather than making you proactive. Also read Mistake Nos. 3, 4, 5 Mistake No. 2: Thinking SPC is a panacea, and will eliminate variation SPC by itself cannot eliminate variation, neither is it a panacea for quality problems. It is one of many in the tool chest and can serve a very useful purpose – that of showing how much variation is naturally inherent to the process, and when special causes occurred if any. Variation occurs because of “Common Causes”, and “Special Causes”, mistaking one for the other leads to serious problems. Consider the case of this pump manufacturer – blaming late deliveries to customers on plant layout. The thinking was that product takes too long to make it through the factory, whereas the real cause was absenteeism / employees reporting for work late on certain days of the month. Over 70,000 dollars were spent changing the plant layout, while the real problem could have been solved for almost no cost by some adjustments in production scheduling, and shift start times. Also read Mistake No. 7. Mistake No. 2 – What You Can Do About It… Setup a good tracking system and be aware that you will need to do more than just SPC. Once causes of variation are understood they will need to be addressed. Invariably there will be two types of causes found – common and special. Some of the causes will be self-obvious, some will require further analysis using simple tools such as Check Sheets, Root Cause Analysis (RCA), etc. However, others will need advanced tools such as Design of Experiments (DOE), Analysis of Variance (ANOVA), Hypothesis Testing, Response Surface Modeling (RSM) to name a few. Variation elimination will come only via removal and / or control of both – common and special causes. In many cases these will become projects that Corrective Action Teams may have to pursue. Management will need to be seriously involved to ensure RCA is done adequately, and corrective action projects get executed. Mistake No. 3: Confusing between Specifications and Control Limits Many times, as data is collected and compared to the Specifications provided by the customer, it is thought that if the outputs are within the Specification Limits, the process is doing fine. Thus, implying that the process needs to be within these limits. Quite often the Specification Limits are plotted on charts and the process is tracked with regard to the same. This is not recommended practice. Specification Limits do not reflect what the process is doing, or is likely to do, therefore do not belong on the Control Chart. Mistake No. 3 – What You Can Do About It… Educate everyone that Control Limits are the Voice of the Process, whereas Specification Limits are the Voice of the Customer. The two are independent, and different entities, and come from different sources. Control Limits are calculated based on the data collected from the process – these are established using Statistical Formulae. When you see Control Charts, make sure to ask how the Control Limits were determined. Mention of anything other than the use of process data (or something to that effect) should be a warning signal. Mistake No. 4: Creating Control Limits based on some one’s judgment, or opinion Several times, as SPC initiatives are kicked off, team members get asked to put certain limits on the chart and refer to them as control limits. When such limits are wide, they give the impression the process is operating in control. The reasons why such control limits may be put in place are many – one of which is fear. This stems from a concern that out of control processes will result in negative consequences or will impair the individual’s track record. Imposing Control Limits on processes cannot make them better, rather it creates misconceptions and deception. Mistake No. 4 – What You Can Do About It… If you are a supervisor, manager, or leader, work to remove the element of fear; otherwise brace yourself to deal with “cooked” numbers and biased charts. These will only “look” good, however will cause enormous harm to the organization in more ways than one. Understand that Control Limits are established based on the data collected from the process and need to have sound Statistical principles underlying the same. Shewhart’s constants have been historically used to calculate Control Limits. If you would like to get a complete list of formulae and constants used in these calculations – please contact me (scroll below to the last paragraph). If you are an employee facing such a situation, you may need to educate others about the basis on which Control Limits are established. One argument you can present is if such limits are decided based on opinion, then, we can have as many control limits as there are opinions. Which limit or limits would be right? Also read Mistake No. 3. Mistake No. 5: Not allowing the process to operate naturally, and tinkering with it constantly When data obtained from the process seem to be unfavorable, an attempt is made to make changes – with the belief that we can improve the process. If the next data point is in the favorable direction, it will reinforce the belief that the change is working. This improvement may or may not last, and then data start going the wrong way. Such situations trigger more aggressive action; perhaps other culprits are chased, and even people fired under the assumption that they messed things up. The cycle continues, with chance variations, and knee jerk reactions – eventually resulting in totally uncontrollable gyrations in the process. Few realize the extent of distrust and morale damage this leads to. The conclusions of what comes next are best left to imagination. Complete overhaul of the plant’s equipment, and or management staff now falls in the realm of possibilities. Those at the receiving end call it “Punish the innocent, reward the guilty”. A similar approach is used when the process is operating favorably as well – changes are made to make it better, and a similar cycle to the one described above gets perpetuated. Mistake No. 5 – What You Can Do About It… First and foremost, understand that one of the underlying assumptions in SPC is that each data point obtained is independent of the previous one. This means that the process must be operating in its natural state and free from constant tampering or adjustments. Only then we can get a sense of how much inherent variation will be present. Making changes as the process is operating nullifies this assumption and begins to introduce bias. The rules of Normal Distribution (or other applicable Distributions) cannot be expected to hold under such situations. Does this mean we cannot make any change what so ever? No. After an acceptable change is made, we must leave the process alone and let it run in its (new) natural state again. Only then we can assess the effects of the change on the process. The process mean will likely exhibit a shift, and once enough data is accumulated new Control Limits may need to be established. Mistake No. 6: Management stays at an arm’s length, and assumes Control Charts will do it all Every now and then, managements realize they need to do something different. An edict is issued to the rest of the organization, with the expectation that the dictums will be obeyed and implemented. This approach rarely works! Theoretically, employees need to listen, understand, and implement the initiatives Management is driving. Today, however, one must “sell” the change. Doing otherwise means use of position power and force, which because of its abrasive nature can create malicious obedience. Once this stage is reached, the initiative is generally doomed. Far too many corporate leaders stay at an arm’s length from the tough work of driving change and working with the troops. Over time, they get out of touch with reality, and live in a world of their own, thinking that SPC is being deployed, as evidenced by Control Charts that are being produced. The only way to detect a change is via examination of the shift in process mean and control limits (assuming the charts are being produced correctly). Just the presence of a control chart does not prove anything. Once it is discovered no improvement has occurred, the program gets buried like many of its predecessor flavors of the month. Mistake No. 6 – What You Can Do About It… Get involved! Establish overall objectives for the program, begin using SPC on key business metrics to set an example. Use Statistical Methods in your decision making and involve your staff in an ongoing dialog on Analytical Methods, including SPC. This is easy to accomplish by setting aside a block of time during your regular staff meetings. More importantly ask key questions about the program, it’s deployment, training of key personnel, and how the success of the program is being measured. Ask for a diagnosis of the causes of variation, and what is being done to address the same. Remember, your staff will follow you, whether you like it or not, therefore leading by example is one of the best ways to make sure your ideas get implemented. Demand more than just creating or looking at charts. Set up aggressive goals for variation reduction, and process improvements. Tie these to customer satisfaction / loyalty, profitability, and implement reward systems based on achievement of such metrics. Involve them in the selection of where to apply SPC and provide extensive training on the topic. Then you will see much easier acceptance by the troops, as well as the rest of the organization. Coach your folks that SPC is not the end, rather a means to an end – one of variation reduction eventually for the betterment of the organization. Mistake No. 7: Mixing up common causes with special causes This is one of the easiest tripping points in SPC. I recall one of the VPs of manufacturing for a leading chip equipment producer insisting that each defect on the equipment shipped by the company be investigated and fixed. This company made complex process equipment, with each machine using over 50,000-piece parts. There were simply too many opportunities for defects, and the effort required to chase every one of them soon got bogged down. While the intent is noble, the means to achieve it were terribly flawed. For any system or organization, current performance reflects: – the way the processes are setup – the policies and procedures in place – decision making practices – training levels of staff – mindsets regarding what is acceptable and what is not (use of Specification Limits) ….and other such elements… Taken together, these determine how the process behaves – by and large, in the absence of rare / special causes. In most cases, there are simply too many systemic causes (common causes) – which makes it difficult to say exactly what caused a given nonconformance. Consider the gas mileage you get from your car. Many variables influence the gas mileage on a routine basis – and the variation may hover within a range of +/-15% (say) from a certain average value. Now, if one day the mileage was found to be lower by 30 or 40% suddenly, it is likely that there was a one of a kind because that resulted in this occurrence. May be there was a football game in town which resulted in massive traffic jams, or a very cold winter day where you had to warm up for a longer time, or maybe something is wrong with the car. On the other hand, if we observe the gas mileage to be lower than average by say just 7%, we would be hard pressed to identify exactly what caused this. Hazarding a guess can result in expensive solutions such as taking in the car for a thorough overhaul, and the outcome may be a hit or a miss. Mistake No. 7 – What You Can Do About It… Start using Control Charts the proper way. First collect data and establish control limits. If violations of control limits are found to occur, it indicates “Special Causes” may be present – these will need to be found and addressed. Then the variation observed from data that lie within Control Limits will be on account of” Common Causes”, or “Systemic Causes”. Such variation cannot be reduced unless the system itself is improved. Improving the system in most cases is beyond the control of process operators, and SPC teams. It is by and large a management function. Making changes to the “System” are much more difficult to accomplish, therefore pushing SPC teams to get these done without supporting them will create frustration. Management involvement in this is key for obtaining results from the deployment and application of SPC. Mistake No. 8: Not checking for data integrity, and measurement system performance Very often these essential and basic steps are ignored altogether. Data that lacks integrity has a devastating effect on any subsequent analysis and decisions. Consider the case of late arrival data from three different airline companies as shown below (values represent minutes late): Getaway Air: 42 15 9 21 12 14 29 10 8 32 22 19 Timely Airways: 6 32 0 48 39 25 14 54 0 0 16 0 Always Flying: 22 0 10 18 19 91 10 14 18 21 10 0 If you were to compute the average late arrival time you will find all have nearly identical performance. A closer examination shows Always Flying has one data point that seems way out of line. Now the question is how should we treat this data point? Keeping it in creates the perception (based on average) that Always Flying is no different from the others. If we decide to eliminate that data point, Always Flying lookg great, however, we can be questioned as to why data was eliminated? Secondly, consider the smallest recorded value in each case. For Timely Airways and Always Flying, it is 0, whereas Getaway air has a smallest value of 6. Further, in the case of Always Flying, the next higher value is 10. Does this point to something? Could it be an internal policy of Always Flying that any delay of less than 10 minutes be considered as the plane having arrived on time? Further, what if the measurement system that has a total error of say +/- 5 minutes? You will not be able to discern any improvements / changes unless a change of 10 or more minutes is observed. Now any improvement smaller than 10 minutes will always be questionable. This has other important implications too. Any data that lie within a 10-minute band from the Control Limits will now be suspect. The control limits thus begin to lose their significance, and ability to detect out of control conditions. Mistake No. 8 – What You Can Do About It… Before you go too far with SPC, lay down ground rules on how the data shall be collected and analyzed. For instance, what will be the least count of the measurement system, and how much error in the system will be acceptable. Often an argument will be presented that the measurement system has been calibrated, and that is the only performance check required. Don’t fall for it. In addition to Calibration, you also need to know the Repeatability and Reproducibility of the measurement system. Other problems with measurement systems include Drift, Linearity, etc. All need to be understood and addressed properly. The total error in Measurement System should not exceed 10% of the “Specification Window”. This will ensure that the Measurement System is crisp, and repeatable. Data integrity needs to include tests of outliers, checks for faulty / contaminated data, validity of data with regard to time and place, etc. The outliers test identifies data points that may be at unjustifiable extreme values; however, this does not license one to automatically throw out such data from the analysis. Outliers need to be examined further to determine if there was an error in recording of data, or an unusual event that generated such data. If an outlier is found to be a genuine data point, you may want to include it in the analysis. Mistake No. 9: Lack of empowerment and trust of teams implementing SPC A common gripe with SPC Teams is that they are getting all this training, but when it comes to applying and living with SPC, management is uncomfortable letting the teams adopt the new methodologies. This is exacerbated by Mistake No. 6; Management Stays at an Arm’s Length. As teams begin putting SPC in place, and allow the processes to operate in their natural state (with some variation), supervisors get quite uncomfortable – because not every unfavorable data point gets addressed any more. This can be very unnerving, since we habitually take action / react for the purpose of maintaining control. What such control yields – is another matter. Mistake No. 9 – What You Can Do About It… Exercise empowerment. If you have provided the right kind of training and mentoring, let the teams work with SPC and allow the rules of SPC to operate. Interfering with the process not only makes matters worse, it sends the wrong message to the teams as well – one that says they don’t count, and what counts is your opinion. If that was to be the case, what was the point in getting the teams up to speed on SPC anyway. Keeping hands off can be hard to do, especially if Managers / Supervisors have been close to processes earlier, and believe they know what needs to be done. Mistake No. 10: Inadequate provision for training and coaching / mentoring SPC is not something one can easily read from a book and expect to implement right away. Lack of training in the application of the tool and driving the initiative can jeopardize the whole effort. Take the case of Mistake No. 8 – Checking Data for Validity and Integrity. Hardly any book on SPC refers to this as a crucial step, and yet – when ignored it can have serious consequences. The few thousand dollars saved by avoiding (or going cheap on) training and mentoring in most cases ends up costing 5 to 10 times more eventually. When training is delivered, but its effectiveness goes un-checked, we do not have any confirmation of whether the participants really learned anything. Mistake No. 10 – What You Can Do About It… Budget for training your associates, operators, SPC teams, professionals, managers and supervisors. Everyone will need to come on board, however the learning needs will be different for different groups. Supervisors and Managers especially will need to understand what SPC is, how it works, how decision making will change after SPC is implemented. They will also need to learn how to implement / drive such initiatives, and most importantly accepting certain level of variation in the process behavior – rather than intervening frequently. Key questions to ask of SPC teams, and operators will be additional topics that can serve Managers and Supervisors well. Operators must know how to avoid mistakes in data collection and handling. They must also know the symptoms as exhibited by control charts to the extent that they have decision making authority regarding process operations. Professionals and SPC teams will benefit from a thorough understanding of the Statistical Principles underlying SPC. Associated topics such as Root Cause Analysis, Process Mapping, etc. should not be ignored, since they often become excellent aids in identifying causes of variation. Also read Mistake No. 5 and 7 Concluding Remarks: I trust you found this article informative. Your feedback is very welcome – e mail me at rai@thekpisystem.com or call: 512-560-8326. You may copy / print / distribute this article as long as the author is acknowledged, and the source of the article is referenced. First published: 2/8/2005 Last updated: 3/24/2022 |
