Pages

Sunday, April 1, 2018

Process Behavior Charts: A Better Way to Evaluate Your KPIs

Have you ever learned something new and thought, ”Gosh! If I had had only known this years ago, my life would have been so much easier!”? We all have, I suspect. What I'm writing about in this post falls into that category.


I remember owning my practice. I dutifully kept stats on everything I could think of:

  • Gross income, 
  • number of new clients, 
  • average invoice total, 
  • number of dentals or spays or neuters, etc. 

I would even plot them on charts and tape them to wall of my office like a “war room,” constantly “watching the numbers” and bouncing emotionally between feelings of “we made it through another period in good shape” and “oh sh*t, we’re down, and this must be the beginning of the end.”


The same sort of thing happened when I worked for a corporate practice. Every week, the practice manager (PM) and I were on a conference call with other PMs, lead doctors and our area managers to discuss “the numbers” -- our KPIs (key performance indicators) -- whether we were achieving our benchmarks, by comparing them to last month, last quarter or last year, and why or, more importantly, why not. Same emotional rollercoaster.


Now, for the stuff I wish I’d known back then. I recently read a book called Understanding Variation: The Key to Managing Chaos 2nd Ed., by Donald J. Wheeler, at the recommendation of my co-blogger Mark Graban. It is a fun little book about some statistics (there is that ‘S’ word) and creating Process Behavior Charts (PBC).


You know that in any process or system there is going to be some amount of variation from period to period. When this is plotted on a chart, it shows up as “ups” and “downs.” Most of this is normal, it is just “noise.” But, sometimes it can mean something significant -- a “signal.” So, how do you tell the difference? By turning your data into PBCs. This way of plotting your data will “filter” out the noise and highlight any “signals.”


I encourage you to read the book as there is more than I can briefly blog about, but, having said that, let me share some points before we get into the charts.


  1. Tables with lists of numbers are difficult to understand. There is no context, and the data is difficult to visualize in this format.
  2. Line graphs, over a longer period of time, are easier to understand and put the data in some form of continuity and context with prior periods.
  3. Comparing two data points, such as the current period data with the same data last month or last year, doesn't offer any context. Who says the data from last period was normal? Maybe, it was a really bad period due to extraneous influences, e.g. inflation or a natural disaster that occurred at that time.
  4. Averages tend to be pretty much in the middle of a range of data. Comparing to averages tends to create “binary output.” You are either above average (“good”), or you are below average (“bad”).
  5. The setting of arbitrary goals, such as a 10% increase over last period, becomes more objective and rational. It is well and good to set the goal, but if the system cannot produce to that degree, it is simply a futile “wish.” No manner of cajoling, incentivizing or threatening employees is going to help. If the goal is outside the limits, then the system is going to have to be changed from what it is right now, and employees have no control over the systems under which they operate. That is management’s domain.
  6. PBCs are the voice of the system. They show how the system is functioning, and the extents to which the system can function, as it is now designed and operating. One can also assume that, without any change, the system will continue into the future as it is currently; it is predictable. If it is not where it should be, then the system has to be changed somehow. It also shows when a data is outside the limits of the system and, therefore, is a signal that something unusual has happened, and it needs to be investigated.


The following data represents the number of new clients seen per month over the last 18 months.


18, 16, 14, 19, 15, 17, 16, 18, 15, 14, 19, 18, 15, 18, 18, 17, 19, 11

Total = 297  Average = 16.5


When just looking at a list of numbers, it is difficult to really appreciate what is going on with the data. In this form, one might easily miss the value of the last data point.

Converting the raw data into a graph is more visually helpful.




You can see that this running graph (or X-chart) is easier to understand and gives better context to the table of numbers.


This appears to be a rather stable system (or process) until, possibly we get to the 18th and last data point. Is this part of the normal “noise” or do we need to investigatte? It is lower than any prior period we have recorded. I can tell you that, for me, this would have been good for at least a week of sleepless nights and two stupid, stress related arguments with my wife!


Continuing with the chart methodology, we next determine the Moving Range (mR), between each two successive data points. This distance is always a positive number, regardless of whether the first number is larger or smaller than the first. For example, the distance between -3 and 2 is 5, or the distance between 8 and 4 (or 4 and 8) is 4.


By comparing the first data to the second, the second data to the third, the third data to fourth, etc., we get the following table:


2, 2, 5, 4, 2, 1, 2, 3,1, 5, 1, 3, 3, 0, 1, 2, 8

Total = 45  Average MR = 2.53



Graphically:





To see if the variation in the X chart is all routine or if there's something exceptional going on in that last data, we then complete the X-chart by calculating and drawing the average, an Upper Control Limit (UCL) and a Lower Control Limit(LCL).  These are calculated as follows:
          Avgx= Totalx/#x
          UCL= Avgx+(2.66×AvgmR)  = 16.5+(2.66×2.53) = 23.54
        LCL= Avgx-(2.66×AvgmR)  = 16.5-(2.66×2.53) = 9.46

Note: The 2.66 is a conversion factor that approximates three standard deviations (but we don't calculate a standard deviation in this methodology).


Updating the X-chart:



To complete the mR chart we need to calculate the Upper Range Limit, as follows:


         Upper Range Limit = AvgmR×3.27  = 2.53×3.27 = 8.66
  
Note: The Lower Range Limit is always zero, since the variances can never be a negative number.


Updating the mR chart:



Together, these two graphs make up an XmR chart or Process Behavior Chart.


Looking at the XmR charts, we can see that the last data point is still within our calculated limits. This indicates that the data is just “noise.” It is just part of the normal variation for this system or process, as it is currently designed.

That said, if we're uphappy with the average level of performance, we could try to improve the system in a systematic way. There's nothing worth investigating in terms of a reactive question like "what went wrong that month?"

The first hint of a "signal" would be any single data point above the upper limit or below the lower limit.


Wheeler’s book gives much more information about the meaning of the charts and some other types of “signals” to be aware of such as: 3 out of 3 or 3 out of 4 data points being closer to one of the limits,

or




a run of eight or more consecutive data points being on one side or the other of the central line are interpreted as being a “signal.”



Mark Graban is currently writing a book on this material. It should be completed by June, but you can buy the first three chapters now through his use of the "Lean Publishing" approach. All of Mark’s books are “Top Class.”


Heads up! I will be the guest on a webinar hosted by Mark Graban and KaiNexus on May 7th at 12:00 noon Central time. We will be discussing the new emergence of Lean management in veterinary medicine.  Use this link. Hope you will join us!

Update 4/4/18: Watch Mark Graban talk about this material here.

Thanks for reading. Tell your friends and colleagues and, as always, comments welcomed.

No comments:

Post a Comment