I still remember the first factory dashboard I helped design. The plant manager wanted a wall of screens showing every KPI you could imagine — OEE, throughput, downtime, energy use, scrap rates, temperature on line three, you name it. So we built it. A massive real-time dashboard pulling from PLCs, MES, and a few sensors we added just for fun. It looked incredible in the conference room.
Then we installed it on the factory floor. Within two weeks, the operators had taped cardboard over the bottom half of the screen. They didn't care about energy use per unit or six-sigma control charts. They cared about one thing: is line four about to jam again, and do I need to clear it now? That was my mistake. I'd built a dashboard for management and put it where operators needed a tool.
Real-time means something different depending on who's standing in front of the screen. For a shift supervisor, five-second refresh is fine. For an operator at a stamping press, five seconds is an eternity. A jam can ruin a die in less than three. We learned to ask what decision the person needs to make in the next thirty seconds, and then we built backwards from there.
We rewrote the operator screens with three big numbers and a color band. Green meant running to takt, yellow meant drift over two percent, red meant stop what you're doing and look. No drill-downs, no filters, no login. They could read it from ten feet away with safety glasses on. The scrap rate dropped nine percent in the first month. Not because we had more data — we actually showed less. We showed the few things that mattered at the exact moment they mattered.
The other thing nobody tells you is that factory networks lie to you during pilots. We tested our dashboard from the IT office on a Tuesday afternoon and the latency was beautiful, maybe 80 milliseconds. Then we went live during third shift. The old WiFi access points near the loading dock started dropping packets every time a forklift drove past with a full pallet of aluminum. The dashboard would freeze, then catch up, and suddenly the "real-time" metric was actually four minutes old. Four minutes. You can't run production on that.
We ended up hardwiring the critical operator stations and adding two access points with industrial enclosures. I should have walked the floor during a real shift before we promised sub-second updates. That walk would have saved us a very awkward plant meeting and one very angry maintenance supervisor.
Now when someone asks me about real-time dashboards for factories, I don't start with data sources or cloud providers. I start with two questions. Who is going to look at this screen, and what bad thing happens if the number is wrong for five seconds? The answers tell you more than any architecture diagram.