Use of Control Charts for Monitoring KPIs with Target Percentages

Six Sigma – iSixSigma Forums General Forums Tools & Templates Use of Control Charts for Monitoring KPIs with Target Percentages

Viewing 8 posts - 1 through 8 (of 8 total)
  • Author
  • #243136


    Hi there,

    I’m wondering is a control chart the appropriate tool to use to monitor KPIs where the data is a percentage, and the target is 100%?

    Since the target is 100%, and a lot of the time that target is hit, the upper control limits are expanding beyond the 100% limit. Should the solution here be to remove upper control limits and only include lower?

    The main goal of using the control chart is to identify a trend. Even though the target is 100%, realistically that target will not be met week after week. Therefore there will be expected variation from the 100% target. So I want to use a control chart and trend analysis to monitor at what point is our performance going beyond the expected variation, and more towards a signal that we are deviating too far from our target.


    Mike Carnell

    @datasciencedweeb If you are using a software package check it to see if you can list 100% as a boundary.



    Nicholas Iacono

    To your first question, yes it is appropriate to use a control chart to monitor KPIs.

    Since you are monitoring a percentage, which control chart to use will depend on how that percentage is calculated (percentages as a metric should throw a yellow flag that you need to do further investigation into that calculation). For example, if the base data behind the percentage is pass/fail, then you’ll be looking at attribute control charts like a p-chart or np-chart. Minitab (if that’s your software) has also added the Laney p’-chart in case your pass/fail data doesn’t fit a binomial distribution (and a p-chart could give you false indications)



    I typically use a Time Series Plot with a reference line for the goal in Minitab.  Since you’re just looking at performance trends you don’t really need control limits  You could also calculate the standard deviation and place a LCL at -3 standard deviations with another reference line.



    I concur with Nicholas that if the percentages are calculated from discrete data, say defectives then you should be using the discrete data (number of defectives and sample size) to calculate the control limits.  In this case when the number of defectives are 0, this corresponds to your 100% case. The control limits for this chart will never exceed 100%. The Laney P’ chart is typically used when your sample sizes are very large.



    Jess L Cotten

    The control charts are good, but you may want to investigate each point of the control limits variance for the true root cause.  This may take you in a different direction in regard to the measure.


    Chris Seider

    Consider what you mean by target of 100%.

    Your control chart would be great if no losses were assumed and the theoretical output was included–not planned output.



    I do use control charts to monitor KPI’s, but I do so reluctantly for lack of a better tool.

    Remember, SPC stands for statistical PROCESS control, not for statistical business control or statistical metrics control.

    The problem is that KPIs involve (mix-up) many different process. Remember: one process makes one product. If you have a machining “process” where you make batches of different parts, each part is done with a different process because the raw materials, machine parameters, work instructions, and product specifications are different for each. Obviously, you would not use the same SPC chart to chart the external diameter of 2 different parts that have different external diameters.

    However, for KPIs one puts all the processes in one bag and call the KPI “scrap” for example, and measure it in scraped parts per week, or % of production that goes to scrap, or $ of scrap per day, etc… But say that one part has a tolerance that is significantly tighter than the other part. Then it is reasonable that the process capabilities will be different for both parts and you will have more scrap in one of them than in the other. Or say for example that you are doing 2 identical parts except one is low carbon steel and the other is titanium, but the % scrap happens to be the same. Well, with the titanium costing more than 10 times what the low carbon steel does, your $ scrap will be very different.

    In those cases, simple mundane things like “this week we receive a large order of part A” and “this other week we decide to build stock of part B in preparation for the season” become special causes of variation (because a process that makes 30% A and 70% B is NOT the same than one that makes 50% each) that can trigger out of control signals EVEN WHEN the processes for part A and B remained perfectly in control all the time. Not very useful, can be misleading, and can lead to chasing false ghosts.

Viewing 8 posts - 1 through 8 (of 8 total)

You must be logged in to reply to this topic.