![](https://static.wixstatic.com/media/346a85_7763c006d80347dcab36215d5d9d2f61~mv2.jpg/v1/fill/w_461,h_612,al_c,q_80,enc_auto/346a85_7763c006d80347dcab36215d5d9d2f61~mv2.jpg)
Do you know what tomorrow will bring? As a manager, can you predict the results of your team’s actions? A Statistical Process Control Chart (SPC), made popular in the 20th century by W. Edwards Deming, is a great tool to help answer that very question. The core idea built into the Control Chart concept is that every system has variation and some of that variation is normal and some of it is not. Before I talk about variation in more dStatisetail, let’s align on the definition of the system. A system implies two things: a common goal and interdependency. In other words, a system unites interdependent functions toward a common goal.
So, what is a “normal” variation? Deming believed that there is variation inherent in any system, in any process or organization. It is there and will always be there. The question is how much normal variation is there in any given process? Furthermore, the whole idea of Continuous Improvement is to reduce, not necessarily eliminate, normal variation, also called “noise”. The goal of a manager becomes the continuous reduction of the noise. To achieve that, the entire system needs to be understood and evaluated since all functions within are interdependent and variation in one component will have an impact on the other.
In addition to the normal variation, there is also a “special” variation – a “signal”. Signals indicate an unusual occurrence and require a closer look. Signals, according to Deming, are what require immediate attention. If a signal is understood and is believed to have a reasonable chance of happening again, it needs to be addressed. How? For negative signals we need to understand the root causes and eliminate or mitigate them. For positive signals we want to replicate them and make them part of normal variation.
Let’s look at the following example (real life data has been changed for the purposes of this demonstration).
![](https://static.wixstatic.com/media/346a85_9bcf0959b1a5459881b745fe5149391b~mv2.jpg/v1/fill/w_980,h_399,al_c,q_85,usm_0.66_1.00_0.01,enc_auto/346a85_9bcf0959b1a5459881b745fe5149391b~mv2.jpg)
The blue line shows company monthly sales over the past five years. The parallel black lines indicate Upper Control Limit (UCL) and Lower Control Limit (LCL) and the red line shows the average of the data. The UCL and LCL are calculated from the data and are three standard deviations from the mean. They are not set to a particular value, like tolerances. They are also often referred to as “natural” process limits. If the sales data were to fluctuate within these natural process limits, it could be said with over 99% confidence level that the sales would continue to fluctuate in the same manner in the near future. What this SPC shows, however, is that there are several signals worth investigating.
Because we used the entire five years to calculate UCL and LCL, as well as the average, we first want to indicate “systemic” shifts when something special had occurred. In this example, we see two periods of interest: one occurring around the end of 2013 and one around the middle of 2016. This means that signals that we observe in 2017 and 2018 (data points that are outside UCL) are for the entire data set and may not necessarily indicate true signals. Upon closer investigation, the dollar increase in sales in both cases (2013 and 2016) was due to price increases!
If we take the shift that happened in 2016 and build a control chart from that point forward, we get a completely different picture.
![](https://static.wixstatic.com/media/346a85_b1ba11fc860e464c9d6646904ec3cc39~mv2.jpg/v1/fill/w_980,h_416,al_c,q_85,usm_0.66_1.00_0.01,enc_auto/346a85_b1ba11fc860e464c9d6646904ec3cc39~mv2.jpg)
The sales, starting in middle of 2016 and going through the end of 2018 show stable and predictable behavior without any signals that we observed in the first chart. This can lead us to believe with a very high degree of confidence that the sales will continue to behave in the same fashion in the foreseeable future unless something major occurs. We could, just like before, increase the price. If that is no longer an option, we have to look at several key matters to see how sales could grow: market penetration, client experience, product quality, etc. The bottom line is that if this company wants to start seeing sales growth, it needs to look at the entire system to find the driver(s). Additionally, we can also assume that what had been done in the past will not deliver desired results considering that previous shifts were due to price increases.
This is an example of building and interpreting a control chart. As you can see, with little effort we can determine what to focus on. Having done that for this particular company we were able to determine that past projects that aimed at increasing sales did not achieve the desired results and something different would have to be done. At this point a more detailed analysis around Client Experience, market data, product data, etc. can be leveraged to see what can be improved.
When we apply SPC to the KPI Framework, we can determine quickly what to react to and how, and what requires a more methodical approach. This can help drastically reduce the number of projects we initiate and change the timing of those initiatives. It can also change which projects we do altogether. If a KPI has a signal, we should investigate. If a KPI shows only noise, but the variation is too broad, we need to look at the system as a whole. SPC gives focus and helps prioritize resources with maximum efficiency.
Comments