Issue link: https://iconnect007.uberflip.com/i/1503998
18 PCB007 MAGAZINE I JULY 2023 activation, perhaps. In that way, you could take some of the temperature, and establish what the estimated three sigma limits are for normal activations. You don't have to explain the math, don't draw them manually; just calculate the limits and report what the pic- ture of normal looks like. We want it to econ- omize our interventions only to when we see very overt signals above or below limits in the activation/deactivation cycle. Only then do we look at the contributory system indica- tions for that. I don't like the idea, though, of leaving lead- ership in the lurch for too long, because they're the ones who need to understand this. Trans- formation to a quality view must begin with them. When leadership is completely over- whelmed by data, they must be able to stand back and say, "Who can explain the most important metrics that we need to know about how our process is working? Because the only real value to me is to know when there are aber- rations in the process, so that we can begin to refine it and we can begin a system improve- ment to reduce that variation." I think Deming had another useful quip, which is a brilliant little inversion: "Don't just do something, stand there." His point was to reduce the incli- nation to tamper with the system straight- away in reaction to some event. Understand what it's doing first, because your reactions to the figures will produce more harm than good; you could ruin things. Where do you see AI fitting in the context of turning this data into knowledge? Is this something that has an application within TQM? I think the jury's still out on that. I've played with ChatGPT in my work, and I find that it guish signals from noise and gain insight into what the data is telling you about how a system or process is working. Here's a real-world example from a par- ticular customer of mine in telecom, where they wanted to track the activation of Apple w a t c h e s a s t h e f i r s t v e r s i o n s r o l l e d o u t . They wanted to know how many people were going onto the network. A senior executive who was very hot to trot on understand- ing this said, "Gather all the data that tells me the activa- tion and deactivation rates of these watches nationwide." It took seven or eight hours a day for people to pull together that data from all the systems and roll it up onto a dashboard as a single num- ber. at number was green when it went up, and red when it went down. You can imagine the attendant behavior which that encouraged. What I suggested to those managers was to present the data in context over time on a run chart with process limits, such that the leadership could see a quarter's worth of Chris Chapman Don't just do something, stand there.