Harnessing the Power of Predictability

Approaching the display of performance data from an S4 point of view can yield much more useful management reports. Consider how a company might monitor a manufacturing process that has specification limits of 72 (minimum) and 78 (maximum). Managers reporting on the performance of this process in a format similar to that of exhibit 2 might be tempted to include horizontal lines on the graph at 72 and 78. Where the process's performance fell below 72 or exceeded 78, they might color the result line red to draw attention to it. But a knee-jerk reaction to these out-of-specification conditions might consume considerable resources and might not fix anything long-term from a process point of view.

Alternatively, when information is presented with a statistical upper control limit (UCL) and lower control limit (LCL), as shown in exhibit 3, at left, report recipients can make statements about common-cause and special-cause variability. The UCL and LCL numbers can be derived easily from the data. Chart out the time-series data points you've collected (represented by the first two columns of exhibit 3's data table), then find the deviation from each data point to the next -- its moving range (MR) value. The UCL is the mean of your data points plus 2.66 times the average MR, where 2.66 is a three-standard-deviation constant that applies when the moving range is determined from adjacent values. The LCL is the mean of your data points minus 2.66 times the average MR.

For control charts that have no data outside the LCL and UCL and show no clear upward or downward trends, quality control professionals will state that the process is "in control." I prefer to simply call the process predictable. What does a process control chart like the one in exhibit 3 predict? We can consider the data in the stable region (between UCL and LCL) to be a sample of not only past process performance, but also future performance -- assuming nothing either positive or negative changes the system. That is to say, we can expect common-cause variability to stay within the bounds of the upper and lower control limits. We can view this data-analysis procedure as a way to get the process to talk to us -- to tell us what we can expect it to deliver in its current state. Upper and lower specification limits should not be shown on a control chart because they have nothing to do with the statement of whether a process is predictable.

Although this example shows a manufacturing process with both upper and lower control limits, the same approach could be used to display data that had only a one-sided specification, such as frequency of on-time departures for an airline, a ticketing quota for a police officer around a particular speed-limit zone, or even desired corporate revenue. The basic point is that if the process is predictable, we need to look at all the data collectively when making a statement about conformance. It's not valuable to try to explain why a single data point is up and another is down. Only if a process exhibits a special-cause condition -- if one data point exceeds the UCL or LCL by a substantial margin or a series of data meets certain other statistical criteria indicating a pattern of unpredictability -- should we talk about data points individually.

Whenever a process is predictable, the next obvious question is what we forecast for the future, based on the data we have. One approach to making such a prediction is to compile all the common-cause data into a dot plot. Since the process is considered predictable, the data from previous periods can be thought of as a random sampling of performance in the future.

Now we insert the specification limits -- the data range that management deems acceptable for this process. A "pass/fail" dot plot like that in exhibit 4, below, highlights the number of data points that fail to conform with the process's specification limits. The 12 percent "defective" rate (i.e., six out of 50 data points) jumps out in exhibit 4. For situations like this, creating a prediction by dividing instances of nonconformance from the total data population is not the best approach. If close-to-specification data points had a slightly higher or lower response, they could have toggled from a conformance to a nonconformance state, or visa versa. And where they end up in the particular cases included in this report directly impacts the predicted failure rate.

A better approach to reporting on this information would be to create a probability plot of all the data from this control chart, as shown in exhibit 5. In this display format, we can see that the process has a nonconformance rate of approximately 13 percent (0.629% + [100%-87.768%] = 12.861%). Because we've already determined through the control chart that any nonconformance is common-cause, we can expect that nonconformance rate to continue in the future unless something either negatively or positively impacts the overall process. If this rate of nonconformance is unsatisfactory, the company needs to modify the basic process to reduce the variability of its outputs. Displaying the results in this probability-plot format shows executives that a process-improvement project, and not one-off firefighting efforts, is the way to reduce the frequency of nonconformance in this process.

Cultural Change Through Reporting

People often talk about wanting to make a cultural change in their organization. Much money is spent on workshops and consulting to address cultural change issues -- and the results are frequently questionable. In contrast, changing the way managers and employees examine and react to performance data can be the impetus for real, long-lasting positive cultural change.

Again consider exhibit 2. If we had drawn horizontal lines at the specification limits and then reacted to each of the individual data points that landed beyond those specification limits, we would be reacting to common cause as though it were special cause. In a culture that responds this way to performance data, one person may receive an award for a long string of outputs that conform to production specifications, while another employee is reprimanded for missing specs a few times in his production work. This can be highly demotivating when the performance differences between the two employees result from random, common-cause variability in the company's processes. It can also lead to poor resource-allocation choices.

Consider the example of an organization that monitors the frequency of accidents. One year the number of accidents causing injury increased during the month of July to 16, up from 14 in July the year before. The company issued a safety memo declaring the increase unacceptable and telling all employees they must watch a 30-minute safety video within the next month. At an average wage of $10 per hour, payroll for the company's 1,500 employees increased by $7,500 for August, not including time spent getting to and from the conference room where the video was shown or time spent issuing memos reminding people to attend, reviewing attendance rosters looking for no-shows, etc.

Safety is, of course, very important. But this company reacted to the increased accident rate without considering first whether the problem was with the process itself. As Deming estimated, 94 percent of nonconforming results are the direct result of the systems management has put in place. The vast majority of instances of poor performance require that the system itself be modified for improvements to occur -- and those modifications require an organization to look at its systems collectively, relative to specification limits, over a long period of time. Reacting to a spike in accidents in an individual month can be counterproductive and expensive.

Organizational performance reporting using individuals control charting (i.e., plotting individual data points rather than mean data), as illustrated in exhibit 3, enables management to view a process in the same detail an airplane would when viewing the earth from flight. We do not end up reacting to small up-and-down terrain differences when flying an airplane. Similarly, in business we should not react to every small instance of lot-to-lot variability caused by differences among people and raw materials. This view from the 30,000-foot level is a no-nonsense methodology that helps organizations distinguish between common- and special-cause process variation so that they can reduce the frustration and expense of firefighting activities.

Cause -- and Effect

Companies do face some performance shortfalls that result from special causes. If an individual problem is determined to be special-cause from a process point of view, the company needs to address whatever was different in that event to cause the nonconforming result (e.g., why was our customer response time yesterday exceptionally large relative to typical common-cause variability?).

Generally speaking, though, organizations can achieve more gains by continuously working to mitigate common-cause problems by improving their basic processes. Effective, long-lasting improvements to processes are not made by firefighting. They require the examination of process data over a period of stability to determine what should be done differently in the long term.

Presenting performance data in traditional management reports, with simple year-to-year comparisons of metrics, may identify results that are out of line with targets, but it does little to help executives determine how to respond to those results. How can a company fix poor performance when it doesn't know what caused that performance? Process improvement projects in Six Sigma utilize a define-measure-analyze-improve-control (DMAIC) road map to investigate the causes behind nonconforming processes using both statistical and nonstatistical techniques. Such an analysis can lead to long-lasting, sustainable improvement, and taking an S4/IEE approach to reporting on the analysis expands the positive impact that companies see in their top-level performance metrics.

Forrest W. Breyfogle III is the founder and CEO of Smarter Solutions Inc. and developer of the Integrated Enterprise Excellence (IEE) business management system. You can reach him at forrest@smartersolutions.com.

Previous 1 2 3