When I first pitched a real-time analytics dashboard to a plant manager, he was thrilled. His exact words were "I want to know the second a machine slows down, not after lunch." Fair enough. We spent six weeks wiring up PLC data, setting up a message broker, and building a screen that updated every five seconds. It looked gorgeous.
Then we put it on the floor and nobody looked at it. For two weeks. The operators kept using their clipboards and the supervisors kept walking the line. The dashboard was a status symbol in the morning meeting, not a tool.
The reason turned out to be simple. We built the dashboard for the plant manager, not for the people standing next to the machines. A five-second refresh felt instant to us in the office, but on the floor it was ancient history. By the time an operator glanced up, the number had already changed three times. They couldn't trust it because they couldn't verify it against what their hands were doing. So they ignored it.
We tore it down and started again. This time I spent three shifts shadowing the setup team on a packaging line. They didn't want a wall of gauges. They wanted two numbers and a trend line: current cycle time versus target cycle time, and downtime in the last hour. That's it. We built that with a two-second refresh and a color change when the line drifted more than 10 percent. Within a week, operators were using it to adjust feeder speeds before the supervisor even noticed.
The technical mistake I made was assuming real-time meant as fast as possible. It doesn't. Real-time on a factory floor means fast enough to act on, and that's a very different thing. A CNC machine's vibration data can be sampled every 10 milliseconds, but nobody needs that on a dashboard. They need a rolling average over 30 seconds that tells them whether the tool is wearing out. Pushing raw millisecond data to a screen just creates noise. And noise becomes background wallpaper.
The other lesson was about ownership. After we rebuilt the dashboard, we left a small panel in the maintenance shop where the shift leads could change the target values themselves. Not through a locked-down admin screen, but a simple plus and minus button. It felt risky at the time. But it made the dashboard theirs. When a new product run started, they updated the target before we even asked. That small bit of trust did more for adoption than any training session.
Now when I design these systems, I start with a question: who is going to change their behavior because of this screen? If I can't name that person and the specific decision they'll make, the dashboard is just expensive decoration. Real-time analytics isn't about speed. It's about closing the gap between seeing something and doing something about it. And on a factory floor, that gap is measured in seconds, not milliseconds.