Just got off a call on a production shortfall that showed up at final assembly and within a few minutes the conversation was locked there. Pull the data, see what changed, figure out what broke. That’s where it always starts. The thing is, nothing looked off. Each site was hitting its numbers, dashboards were green, if you just scanned it you’d say the system was fine. This one was spread across three plants: component coming out of one site, subassembly in another, final build somewhere else, everything moving on a tight cadence that doesn’t leave much room for drift.
In PowerBI it all lined up, standard processes, expected outputs, quality thresholds, governance, from the corporate view it holds together. But of course, where this usually breaks down is in how things move between sites. A couple of days after the issue first showed up an analyst had pulled a small detail that hadn’t shown up anywhere else, slight dip in yield at the component site, low single digits, nothing that would trigger escalation on its own, and it had been sitting there for about a week.
That was it. It never surfaced in a way that connected back to final output. Each site was still “performing” and it still didn’t come together. At that point everyone’s trying to piece it together after the fact trying to figure out when it started and how long it had been building. The data’s there it just doesn’t connect in a way that helps when you actually need it.
“Trust but verify” comes up a lot in these conversations (that’s how HQ runs the business – trusting that the sites are watching/fixing issues all the time,) but the verify part just doesn’t travel well across a network like this. The situations that run better are the ones where you can *actually see* how work is moving across sites, not just the outputs but the flow and how upstream and downstream are tied together.
When something starts to slip, yield loss, schedule misses, materials moving between locations, you can see it forming instead of hearing about it at the end. It usually comes down to a handful of signals that reflect the system, not each site in isolation. Once that’s in place the conversation shifts pretty quickly, people stop staring at one location and start asking where the constraint actually is.
More tech and AI are getting layered into these environments right now and it’s making this more obvious. But you can keep improving individual nodes and still feel like the whole thing is off! The work is in seeing how it actually comes together, and where it starts to separate.