A vibration sensor on a crusher reads 18% above normal.
Is that a problem?
It depends. Was the bearing replaced last week? Is the plant running a harder ore blend this shift? Did the same reading show up before the last failure, or before nothing at all? Is the line ahead of plan, or already behind?
The number is accurate. On its own, it's almost useless. The decision depends on everything around it.
An experienced supervisor fills in that picture without thinking. Most AI systems can't. And as industrial AI moves from pilots into daily operations, that gap is becoming the difference between AI that helps run a plant and AI that produces reports nobody acts on.
The shift is already underway. In Rockwell Automation's 2026 State of Smart Manufacturing report, 59% of manufacturers said they now actively use smart manufacturing technologies in day-to-day operations, while only 18% are still piloting (Rockwell Automation).
Gartner expects the move to accelerate. Its 2026 Magic Quadrant for Global Industrial AIoT Platforms predicts that by 2030, 65% of AI agents from industrial AIoT platform providers will be in production across asset-intensive industries, up from 3% today (as quoted by ABB).
That's a lot of AI about to make recommendations, and eventually decisions, on real equipment. The question is what it will base them on.
Most plants don't lack data. They have plenty, spread across machines, historians, MES, ERP, maintenance systems and the quality spreadsheet someone updates every Friday.
Each source can be accurate on its own and still mislead when it's read without the others. A downtime event in the historian doesn't say which order was running. The work order in maintenance doesn't say what the line was producing when the fault started. The ERP knows the order was late, but not why.
Giving AI access to all of it doesn't fix that. Connecting systems moves data from one place to another. Context is what tells you how the pieces relate: this machine, during this order, on this shift, right after this change.
The industry is converging on the same conclusion. After interviewing more than 30 vendors at IMTS 2026, IoT Analytics reported that companies increasingly see the context layer as their main differentiator, and advised manufacturers to look past "AI-ready" claims and test what sits underneath (IoT Analytics).
Much of the industry conversation about context focuses on product and engineering data: revisions, configurations, bills of materials. That matters. But on the plant floor, the context that drives everyday decisions is operational. It comes in four kinds:
An experienced supervisor holds all four in their head. AI needs them supplied, every time, automatically.
Take the crusher from the beginning. This is an illustrative example, but every mine will recognize it.
Without context, the system raises a vibration alert. It joins the other alerts from that shift. A maintenance planner sees it the next morning, schedules an inspection for later in the week, and moves on.
With context, the system knows the bearing was replaced six days ago, that the same pattern appeared twice before unplanned stops in the past year, that the current ore blend is harder than usual, and that a stop in the next 12 hours would hold up the mill feeding a priority shipment. It doesn't just alert. It ranks this above everything else on the shift, explains why, and recommends checking the new bearing's alignment before the end of the shift.
Same sensor. Same reading. Completely different outcome.
None of this requires replacing the systems a plant already runs. It requires connecting them in a way that preserves the relationships between them, then putting that picture in front of the people, and the AI, that make decisions. At B3, we do it in three steps:
The order matters. AI is only as good as the picture it reasons over. Build the context first, and the AI on top of it starts making calls an experienced operator would recognize as right.
The next wave of industrial AI won't be decided by who has the smartest model. It will be decided by whose AI understands the operation it's working in.
See B3 in action · Calculate your downtime cost
It's the information that explains a data point: which asset it came from, what was being produced, what happened just before, and what it means for cost and priorities. It turns a raw reading into something a person or an AI system can act on.
Connecting systems moves data into one place, but it doesn't automatically capture how the pieces relate in time and across assets. Without those relationships, AI can read every number and still misjudge what matters.
Product context describes what is being made: designs, revisions, configurations, bills of materials. Operational context describes what is happening on the floor right now: the order running, the state of each asset, recent changes and the cost of a stop. Industrial AI needs both, and most plant-floor decisions depend on the second.
No. A connectivity layer like Orion links the machines, historians, MES, ERP and maintenance systems you already run, so context is built on top of your current stack.