Three months into a smart factory pilot, I stood on the plant floor and watched exactly zero operators glance at our new real-time dashboard. It was mounted on a huge monitor right above the line, pulsing with OEE numbers, cycle times, and alerts. But nobody cared. They walked past it like it was a screensaver. I had pulled all-nighters getting that data pipeline to update every 200 milliseconds. Turns out, speed was the least of our problems.
My biggest mistake was designing the dashboard for the plant manager, not for the people actually running the machines. I remember sitting in an air-conditioned conference room with the ops director, picking KPIs and color schemes. He wanted to see aggregate throughput, waste percentages, and a trend line for the shift. Fair enough. But what shows up on a big screen when you're five feet away needs to be legible and instantly meaningful to someone who's been on their feet for six hours. Tiny font, six gauges, and a scrolling feed of alerts just became visual noise.
What I didn't consider was that operators already had a real-time dashboard: their own senses. A subtle change in machine vibration, a slight delay in part ejection, the smell of oil getting too hot. Those cues happen at the machine, in the moment. A screen six meters away can't compete unless it tells them something they can't feel themselves, like a looming downstream bottleneck or a quality trend drifting out of spec before the CMM catches it. We were showing them what they already knew, just a few seconds later and with nicer graphics.
The turning point came when I spent a full shift shadowing a line lead named Maria. She didn't mince words. She pointed at the giant screen and asked, "Is that for me or for your meetings?" I tried to explain the OEE breakdown, and she just shook her head. "I need to know if the labeler is about to jam, not some percentage." That hit hard. I went back and scrapped the entire UI. We replaced the cluttered dashboard with three huge numbers: current units per hour versus target, the age of the oldest unreleased batch past the QC window, and a simple green/yellow/red status for each station. Underneath, we kept all the detailed analytics for the supervisors' tablets, but on the floor, it had to be glanceable.
We also moved the screen. Instead of above the line, we put monitors at each station with just that station's status and a timer counting down the next PM window. That's when things clicked. Operators started looking. Some even gave feedback when the color thresholds felt off, which meant they trusted it. One guy taped a sticky note to his monitor reminding himself to reset the counter after a die change. That's the kind of engagement you can't design from a conference room.
The experience changed how I approach factory analytics now. Real-time doesn't just mean low latency; it means matching the pace of the worker's day. Data has to land at the right point of decision, not just at the right millisecond. I'll still build the fancy models and data lakes, but before I touch a frontend, I ask a line worker what one piece of information would make their next hour easier. Usually, it's not a dashboard at all. Sometimes it's just a simple light.