How “Process Behavior Charts” Save Us Time and Help Us Sleep Better at Night
I've beenwriting blog posts about using “process behavior charts” – a form of “Statistical Process Control (SPC) chart” or “control chart” that can be very useful to leaders as a way to better manage our business metrics and performance measures. I have also been teaching a作坊在这种方法上。
EDIT:See the recordingof my talk from Lean Startup Week.
There are times when I get the opportunity to coachthe KaiNexus teamon these approaches. These are methods that can be used in manufacturing, healthcare, startups, or any setting where we track metrics and want to improve.
My ears perk up when I hear people like我们的营销总监新利18网址（谁是惊人的）在Web会议期间说：
It's low. I don't want it to be low.”
The number of new contacts for a company perhaps isn't the most vain of the “vanity metrics” (as they say in The Lean Startup movement – seethis blog post by Eric Ries). Some contacts turn into “leads,” some of those become “qualified leads,” and some of them will become paying customers (a metric that matters for the long term).
Vanity Metrics: “They might make you feel good, but they don't offer clear guidance for what to do.” – Eric Ries
当玛吉表示关注时，我听到了她谈论一个数据点（本月），那点低于平均水平或低于预期或欲望。If a number is lower than the previous month, is that really worth reacting to?Is that data point a “signal” (or part of a signal) or is it “noise?”
How I Determined This
During our web meeting (it was just the two of us), I suggested that we look at the chart to see a longer trend than two data points. We use ourKainexus.software platform to track our internal improvement work and our metrics, so we thankfully had a run chart to pull up and view together. Here is that chart, but I've covered the Y-axis numbers:
The good news is that the number of leads appears to be increasing over time. It doesn't look like a “stable and predictable” system with flat performance and results that fluctuate around a consistent average, like this following chart that I used in本文。
In that sort of system, we'd learn to recognize that the results are going to be higher some months than others. Some will be above average and some will be below average. We'd expect that, unless something significantly changes in the system that patient satisfaction in the next month would be between 80 and 92 percent based on our calculated upper and lower control limits.
看看我的book “成功措施” to learn about this methodology and “process behavior charts”:
I downloaded the “new contacts” data set from KaiNexus to create a control chart (aka a “process behavior chart”) in Excel. I calculated an average and the control limits based on the first 20 data points and got this chart with some calculated upper and lower limits (these limits are calculated based on the “moving ranges” between consecutive data points –learn more here).
The chart confirms (more mathematically than me just eyeballing it) that the number of new contacts is not a flat, stable, and predictable system over time. We see a data point around 7/26/15 that's above the upper limit and we see a run of far more than eight consecutive data points above the established average (the green line) – these are bothindicators“特殊原因”，这意味着它们在统计上不太可能是噪声或正常变化的结果。我们可以问“发生了什么事？”在这些情况下，我们可能更有可能找到解释或根本原因。
The next question in my mind is “is the system increasingly linearly?” or “do we see a series of step function increases in performance?” While we talk about continuous improvement, the reality might be there were just a few significant changes to the system as opposed to continued changes and improvements each month.
Instead of being satisfied with the linear trend line, I looked for statistically significant shifts (step function changes) in performance, where the average and control limits shift.
Long story short, I ended up with a chart, below, that I think better reflects reality.
The process behavior chart suggests that there are three different systems over time. There appear to be two step-function increases in the average number of contacts being collected each month. What happened to shift the average performance up? We're not really sure, unfortunately. Had we been using the process behavior chart at the time and detected those signals, it would have been easier, perhaps, to develop a theory about why it improved. Charts like these tell us there's a signal – it doesn't tell us why (that's up to us as the users of the charts).
There also appears to be a shift up in December 2016, with five consecutive points that are above the “1 sigma” limit (another one of the西部电动规则）并且在旧的平均水平上方有八点（其中两个略低于平均值 - 我可能一直在审判这一规则）。
无论哪种方式,它似乎是一个平坦的系统和fluctuating around what I'll call that third average. I don't see evidence that it's continually increasing every month like a linear trend line.
回到玛吉的原始问题——的数量new contacts seemed low to her through September 20. We extrapolated that number for the entire month (multiplying it by 1.5) and that number was between the lower limit and the average.
Maggie said, “So I shouldn't lose any sleep over September's number?”
那说，如果她对当前平均联系人数量不满意，she can do things to improve the system。不要试图通过询问更改的东西（当它可能没有变化）进行改变的情况而不是试图解释9月的号码，而不是尝试实验并花时间做可能会增加联系人数量的事情（例如写一个额外的电子书）。
Does your organization ask you to over explain every up and down in performance measures? Would that time spent cooking up an explanation be better spent improving the system?