{"id":7530,"date":"2018-08-14T00:00:00","date_gmt":"2018-08-14T00:00:00","guid":{"rendered":"https:\/\/lei.flywheelsites.com\/the-lean-post\/articles\/react-less-and-improve-more-by-using-spc-more-effectively\/"},"modified":"2021-10-06T22:35:11","modified_gmt":"2021-10-07T02:35:11","slug":"react-less-and-improve-more-by-using-spc-more-effectively","status":"publish","type":"post","link":"https:\/\/www.lean.org\/the-lean-post\/articles\/react-less-and-improve-more-by-using-spc-more-effectively\/","title":{"rendered":"React Less and Improve More by Using SPC More Effectively"},"content":{"rendered":"\n

Lean transformation work is difficult and time consuming. Nobody complains about having too much time in the day. Therefore, we need to prioritize our use of methods like A3 problem solving and other ways of grasping the situation. Managers and change agents often err in reacting to insignificant changes in a metric. Or, they declare victory too quickly when a single data point suggests improvement. As I explain in my new book, Measures of Success<\/em><\/a>, when we stop reacting to \u201cnoise\u201d in a metric, we can better focus our improvement efforts, waste less time, and improve more. That should be a core principle of Lean management that can help leaders in any setting.<\/p>\n\n\n\n

While the heart of most improvement work is based on tracking and responding to meaningful metrics, I\u2019ve seen far too many lean efforts fall short due to confusion about metrics and KPIs. And not simply because people use the wrong measures. What\u2019s worse is that people tend to apply misguided thinking about mining for gold from the cache of data. Donald J. Wheeler, Ph.D., the author of the fantastic book Understanding Variation<\/em>, says that Statistical Process Control (SPC) is \u201ca way of thinking, with some tools attached.\u201d That reminds me of what people say about Lean — that it\u2019s not just tools, but also a new way of thinking.<\/p>\n\n\n\n

Lean thinkers should recognize that SPC charts broadcast the \u201cvoice of the process\u201d by creating a language of improvement that calls out meaningful signals that our system has changed significantly. This vocabulary only makes sense, of course, to those who are willing to listen — to investigate data without bias, in a holistic manner, with a willingness to identify and respond to signals, instead of overreacting to every bit of noise in the metric.<\/p>\n\n\n\n

Do you use SPC concepts as part of your organization\u2019s Lean journey? Have you discovered this proven, simple methodology to evaluate your metrics and Key Performance Indicators? If not, you might be wasting a lot of time asking people to investigate and react to meaningless changes in a metric. We\u2019re supposed to be opposed to waste, right?<\/p>\n\n\n\n

When I started working at General Motors in 1995, we used SPC as a tool. We had SPC charts on the shop floor before we later started emulating Toyota\u2019s \u201cLean\u201d practices. Production workers gauged parts and plotted key engine block dimensions dutifully on an SPC chart. However, our senior leaders couldn\u2019t see what the charts were really telling us. For example,  they refused to let workers stop the line when the SPC chart went \u201cout of control\u201d, saying that the parts were still in spec. They failed to recognize that those moments were lost opportunities to learn and improve; and workers got understandably frustrated (and customers ultimately suffered) when out-of-spec parts later were delivered to the final assembly plant. Workers used the tool, but management didn\u2019t share the way of thinking. That was an unhelpful combination.<\/p>\n\n\n\n

What Would Toyota Do?<\/strong><\/p>\n\n\n\n

A few years back, I was part of a group that toured the Toyota truck plant (TMMTX) in San Antonio. A Lean Six Sigma Master Black Belt who was part of our group asked our guide, \u201cDoes Toyota use Six Sigma?\u201d The guide said, \u201cNo, but we teach everybody the seven basic quality tools<\/a>.\u201d These tools, which gained popularity during the Total Quality Management (TQM) era, include methods that many might label now as \u201cSix Sigma\u201d \u2013 including SPC charts, a.k.a. \u201ccontrol charts.\u201d<\/p>\n\n\n\n

In The Toyota Way Fieldbook<\/em><\/a>, Jeff Liker and David Meier wrote that companies should \u201cconsistently measure and plot the measures on simple visual trend charts.\u201d The book doesn\u2019t discuss SPC, but instead shows an example of a \u201ctrend chart\u201d (a.k.a. a \u201crun chart\u201d) comparing performance over time against a goal. It\u2019s important to compare our performance against a goal, but we should also ask another important question: \u201cAre we improving?\u201d<\/p>\n\n\n\n

A \u201cProcess Behavior Chart\u201d (Wheeler\u2019s term for a specific form of SPC chart) provides a great way to see if we are improving. A Process Behavior Chart has the added benefit of helping us avoid overreacting to routine variation (\u201cnoise\u201d) in the metric that\u2019s not indicative of a change in the system that produces those results. We\u2019ll see examples of this below.<\/p>\n\n\n\n

We Can Improve the Way We Measure Improvement<\/strong><\/p>\n\n\n\n

A Process Behavior Chart (or even a simple run chart) provides more context than the methods that I most commonly see embedded in modern-day \u201cLean Daily Management\u201d practices in healthcare and other settings.<\/p>\n\n\n\n

Look at Trends Instead of Making Simple Comparisons to Goals<\/strong><\/p>\n\n\n\n

I often see a leader posting a single data point that represents the most recent day, week, or month. Quite often, this data point is compared to a target and the number is color coded as red (not meeting target) or green (being better than the target). That color coding is, in a way, visual, but doesn\u2019t allow anybody to see trends over time.  <\/p>\n\n\n\n

\"\"\/<\/figure><\/div>\n\n\n\n

<\/span> <\/p>\n\n\n\n

A leader might also compare the most recent number to the previous period\u2019s number, saying that it is \u201cup by X\u201d or \u201cup X%.\u201d However, two data points are not a trend. The factual statement about a percentage improvement doesn\u2019t tell us if that change is signal or noise. A Process Behavior Chart can do that.<\/p>\n\n\n\n

Avoid Striking Out with Bowling Charts<\/strong><\/p>\n\n\n\n

I also see many Lean Daily Management boards that display so-called \u201cbowling charts\u201d \u2013 dense tables of numbers, usually with color coding. It\u2019s really difficult to see trends in such visual displays, as seen below.<\/p>\n\n\n\n

\"\"\/<\/figure><\/div>\n\n\n\n

<\/span><\/p>\n\n\n\n

Looking at the bottom metric, the patient hospital rating, we see a mix of red and green. Are we meeting the target? Sometimes. Are we getting better? Can we predict future performance? Should we react to any of the latest figures? Those questions are harder to answer from the bowling chart. We might have been excited by three consecutive greens, only to be discouraged by the reds that followed. But, are all of those red data points worth reacting to? A Process Behavior Chart can point the way.<\/p>\n\n\n\n

Use Process Behavior Charts to Know When the System Has Changed<\/strong><\/p>\n\n\n\n

Process Behavior Charts allow us to make better management decisions that lead to less wasted time and more improvement. When we react less, we can focus our efforts and improve more.<\/p>\n\n\n\n

If the hospital ratings metric (the bottom row of the bowling chart) is visualized as what\u2019s called an \u201cX Chart,\u201d we see the following chart, which shows the data, the average, and two lines that represent the \u201cNatural Process Limits\u201d of the system and the metric. These upper and lower limits are calculated from the inherent level of variation in our baseline data. These limits are not chosen and should not be confused with targets or goals.<\/p>\n\n\n\n

\"\"\/<\/figure><\/div>\n\n\n\n

<\/span><\/p>\n\n\n\n

The X Chart makes it much clearer that we had a metric that had been fluctuating around an average that happened to be near the target. There is routine variation, or \u201cnoise,\u201d in any system or metric. A stable workplace system will sometimes generate a higher number in some months and a lower number in others. That\u2019s a fact of life. Not every change in the metric is equally meaningful.<\/p>\n\n\n\n

The last data point, which might have looked like just another red box on the Bowling Chart, stands out on the X Chart. That last data point is well below the calculated \u201cLower Natural Process Limit.\u201d The last data point is a \u201csignal\u201d that something has significantly changed in the system and is worth investigating \u2013 we could grab an A3 and do some root cause analysis, trying to understand why the hospital rating score dropped so much, so quickly. The August data point is actually just below the lower limit, so that would have been an even more timely time to react and that could have been an early indicator that our process was degrading before the big drop in September<\/p>\n\n\n\n

The Process Behavior Chart is designed to filter out noise in the metric so we can detect signals. In too many \u201cLean Management System\u201d environments, I see leaders asking people to explain noise in the metric. There is no \u201croot cause\u201d to be found for every up and down in the metric. Noise is generated by dozens or sometimes hundreds of \u201ccommon causes\u201d that are commingled and intertwined.<\/p>\n\n\n\n

It would not be a good use of time for a leader to ask, \u201cWhy was the July number lower than June?\u201d The answer to improving the metric would not be found in asking, \u201cWhy was that one month better?\u201d if that month was also part of the noise in the metric.<\/p>\n\n\n\n

Three Rules for Finding Signal in the Noise<\/strong><\/p>\n\n\n\n

The Process Behavior Chart methodology gives us three rules to use when evaluating an X Chart for signals. We don\u2019t have to guess. We don\u2019t have to rely on rules of thumb like, \u201cInvestigate every below-average data point\u201d or \u201creact when we have two data points in the red.\u201d Instead, you might consider reacting to any of these three rules:<\/p>\n\n\n\n