Industrial AI isn’t failing because of data or models. It’s failing because most systems can’t operate at the moment of decision. At AFPM, that gap between ambition and execution was impossible to ignore.
At the AFPM Annual Meeting this year, our team watched a room full of executives answer the same live poll questions (Slido) and arrive at completely different conclusions. It was the same data, but the realities were different.
The responses revealed a deeper disconnect.
Executives are aligned on the promise.
Operators are clear on where value actually shows up, at the moment of decision.
But the two are not aligned on how to get there.
That gap, between ambition and execution, is becoming the defining challenge of industrial AI.
And it’s widening.
Digital transformation has dominated boardroom conversations and conference themes for years.
AI is now positioned as the next competitive lever — one that promises:
Leadership expectations are clear and increasingly measurable.
Executives aren’t asking “Can AI work?” anymore.
They’re asking: “Can you prove ROI tied to margin — and can you scale it across sites?”
That’s a very different bar.
But inside the plant, the conversation sounds very different.
Operators aren’t really asking for transformation. They are asking for something much more immediate, much more practical:
“Can you help me make a better decision right now, on this shift, at this moment?”
A common objection we hear from organizations is that they don't think they have the data, or data in the right place or the right format. In reality, this is not a data problem.
This is a data access problem - frequently it does exist, but it is scattered and sometimes fragmented. And that is eating people's time, obscuring their ability to extract the data from the context of how it's presented, and ultimately creating a cognitive load disaster.
That makes it a timing problem. A context problem. A cognitive load problem.
Because in reality, in the day-to-day, the frontline is operating inside constraints that most transformation narratives ignore.
Their questions are grounded in what’s happening in front of them, in the moment.
And if we’re honest, the answer right now is often no. In one discussion, it surfaced that people are spending 20% of their time just searching for information.
Pause on that. That isn’t just inefficiency.
That is lost attention during moments that require clarity.
It is tempting to simply label this as an AI gap.
Or a workflow issue. It’s both.
And more importantly, it’s the intersection of where things are breaking down. The Slido data made the tension visible.
Slido Question: Leadership’s primary expectation from AI:
Answer: Productivity gains and safer operations (~51%)
Slido Question: But where people actually feel AI making a difference:
Answer: Improving operating decision-making
These are connected, but they are not the same.
One is strategic. Measured over quarters, across sites, in aggregate outcomes.
The other is tactical. Felt in seconds, inside a single decision, under pressure.
And here is the gap:
You do not get the strategic outcome without solving the tactical reality first.
Until AI shows up at the exact moment an operator needs it, transformation remains theoretical.
Because for the frontline, value isn’t defined by the dashboards, models or long-term potential.
It’s defined by one question: “Did this help me make a better decision right now?”
Most AI systems weren’t designed for this kind of environment. Not because they lack sophistication, but because they were built on the wrong abstraction.
They model individual units, generate isolated insights, and depend on workflows that sit outside the moment of action. But a real plant doesn’t operate that way. When a crude unit is pushing for margin, a downstream unit is approaching its constraint, and a shift change is happening simultaneously, these aren’t separate problems. They are one system-level decision that needs to be resolved in real time.
That mismatch is what shows up in practice: static models degrading, dashboards being ignored, and pilots stalling.
The issue isn’t that operators lack information. It’s that the information exists in silos that are optimized independently, handed off sequentially, and never reconciled in real time. That’s what breaks down. And that’s what the industry is finally starting to name.
This breakdown isn’t theoretical. It shows up clearly in what companies report:
These aren’t isolated issues. They’re different expressions of the same underlying problem: systems built to optimize parts, not coordinate the whole.
This shows up in statistics like these:
The industry isn’t failing at AI.
It’s failing at operationalizing AI. Because most systems still sit outside the moment of action, rather than operating inside it.
.png)
Backdrop: A Market That Punishes Delay
All of this is happening against a volatile global backdrop.
Energy markets are tightening.
Supply chains are shifting.
Geopolitical risk is redefining profitability windows.
Margins are no longer stable.
They’re fleeting.
Which means: Decisions matter more — and they matter faster.
So the question becomes:
How do you bridge the gap between the “Ivory Tower… and District 12”?
Between:
Between:
Closing the gap requires more than better models. It requires systems built to operate inside the decision loop itself.
LLMs have spread into the public domain at unprecedented speed, raising real concerns about opaque reasoning, IP leakage, and loss of control over how knowledge is used and interpreted.
In industrial settings, their role is much narrower and more grounded; serving as an interface to industrial knowledge.
One capability that is already delivering value today: LLM as a new interface to industrial knowledge.
They are not the transformation itself. But have the ability to remove a major source of friction. Operators will no longer have to hunt across systems, documents, and tribal knowledge to answer basic but critical questions:
Instead, they would ask and get answers in context. This makes a dent in the 20% of time searching for information, LLMs compress that dramatically.
They can:
This improves productivity. But more importantly, sets a foundation for better in-the-moment decisions. But this alone doesn’t change how decisions are executed. That’s where most systems fall short.
The second category is where most AI efforts break down. It's built on your own data, inside your own environment, and tightly coupled to how your plant runs.
Because of that, it doesn’t generalize. And it isn’t meant to.
Improving insights is not the same as improving outcomes.
In a real plant, decisions are not made in isolation. They are:
And yet, most AI systems still behave as if a single recommendation is enough.
It isn’t.
What’s emerging instead are systems that operate inside the decision loop, not outside it.
Systems that:
For example:
A crude unit pushes throughput to capture margin.
At the same time:
A traditional system flags each of these independently. An operations-focused system does something different: It resolves them together, and guides a plant towards the best coordinated action in real time.
That’s the shift.
Not better dashboards.
Not faster alerts.
Coordinated Operating Strategy: one that reconciles every constraint, every unit, every role, in real time and keeps the human in the loop where it matters.
That is exactly what the AFPM audience signaled:
The signal is clear:
The future is moving past “What should I do?”
Toward: “How do we execute the best decision, consistently, across the system?”
A Necessary Reality Check
But let’s be clear about one thing.
Fully autonomous plants?
Still a myth.
Even the industry acknowledges:
Removing the human in the loop is a pipe dream.
The goal isn’t replacement.
It’s amplification.
The companies making real progress share one thing: they stopped treating AI as a technology initiative and started treating it as an operating model change. That means digital and operations aren't separate workstreams. It means success metrics are tied to margin and throughput and not model accuracy. And this translates to not just going from pilot to pilot but a proof point designed to scale.
AI solutions are only solutions if they scale.
Industrial AI is not early anymore. But it’s not fully mature either.
Regardless of how organizations view themselves today:
There is a gap.
The companies that win the next phase of industrial AI won't win because they have the most sophisticated models. They'll have the clearest operating picture and a system that helps every role act on it, consistently, shift after shift. That's the real gap that we see. The good news? It's a very closable gap and we've seen it first hand.
Because in industrial AI —
The future isn’t about intelligence.
It’s about execution.
