The OT/IT Divide: Why Industrial Data Stayed Trapped for Decades
Operational data spent decades locked in the systems that produced it. The story of the OT/IT divide: why it existed, and what finally made plant data reachable.
For most of my career, the hardest part of working with industrial data wasn't generating it. A plant generates an overwhelming amount of data — every sensor, every loop, every trip, logged to the millisecond. The hard part was that the data stayed where it was born. It lived in the PLC, in the historian, in a SCADA tag database, in formats and protocols built for control, not for analysis.
That made sense for a long time. Control systems are designed to be deterministic and closed. You don't want your safety logic reaching out to the internet. But it also meant that the people who could most benefit from operational data — process engineers, reliability teams, and now AI systems — often couldn't reach it without a painful, custom integration each time.
The seam between two worlds
What's changed is that the tools for moving data across that seam have finally matured. OPC UA gives a standard way to expose tags. Lightweight services can subscribe to those tags and stream them somewhere durable. Document and time-series stores can hold years of plant history cheaply. And retrieval systems can make all of it queryable in plain language. The shape of that pipeline, from PLC to cloud, is what I lay out in my OT to cloud architecture guide.
The value isn't in the data. It's in the distance you can move it without breaking the thing that produced it.
The discipline is in not breaking anything on the way. A pipeline that pulls from the historian must never put load on the control network. Read paths and write paths have to stay separate. The data leaving the plant has to be the data — not a lossy summary that quietly drops the events you'll later wish you had.
Why this matters now
Once operational data is somewhere open, the questions you can ask of it change completely. Instead of an engineer manually pulling trends after a trip, an agent grounded in that data can answer "why did line 3 stop?" against the actual history. That's not magic — it's a retrieval problem on top of a data-movement problem. Both are solvable. Most of the work is in doing it safely. The concrete, no-historian version of that data-movement problem I walk through in monitoring industrial data without an expensive historian.
Written by Usman Nasir — control systems engineer, Stockholm.