Stephen Fowler, Senior Advisor, Applied Computing

For most of my 35-year career in manufacturing, I have seen the industry evolve dramatically — but struggle with one persistent challenge.

We have invested in better processes, more sophisticated control systems, and increasingly powerful digital tools. We have improved reliability, strengthened governance and optimised performance in individual areas. Yet asset management has remained largely reactive.

After every major upset, we would uncover the same conclusion: the signals were there. A gradual decline in efficiency. A vibration trend creeping upward. A constraint emerging between upstream and downstream units. But the connections were only clear after the event.

What is changing now is not simply the rise of artificial intelligence. It is the way AI can be architected across an asset. In my view, applied computing’s Orbital platform represents a best-in-class example of how this architectural shift could define the future of energy operations.

Looking back explains why.

From limited visibility to fragmented optimisation:

In the 1990s, the constraint was limited data. Engineers gathered readings manually, assembled reports painstakingly, and made decisions with partial visibility.

By the 2000s, discipline improved. Maintenance execution, equipment integrity frameworks and structured governance strengthened performance. But these improvements also created silos. Each function optimised within its own boundaries.

By the 2010s, the constraint reversed. Data was abundant. AI pilots and digital initiatives promised transformation. Yet many solutions were bespoke, resource-intensive and difficult to scale.

We improved components of the system — but rarely the system itself.

The 2020s: why AI didn’t transform everything overnight:

In the early 2020s, AI capabilities accelerated rapidly. Yet heavy industry did not immediately transform.

There are clear reasons for this. Industrial environments are safety-critical and capital-intensive. Leaders cannot deploy unproven tools that risk reliability. Early AI efforts often required significant data preparation, custom integration and specialist expertise. Many tools generated insights, but did not embed directly into operational decision-making.

The limitation was not ambition. It was architectural fragmentation.

The cost of hindsight:

A recurring theme persisted across decades. After every major upset or sustained performance gap, investigations concluded with the same phrase: “with the benefit of hindsight”.

The interactions between units were visible in retrospect. The opportunity losses were measurable. Leadership teams could quantify “money left on the table” at month-end — but could not recover it.

The challenge was not effort or expertise. It was the absence of a system-level intelligence layer capable of connecting signals dynamically across disciplines in real time.

Why architecture matters:

Industrial assets are complex, interconnected systems. A reformer affects hydrogen production. Hydrogen availability influences hydrotreaters. Equipment degradation impacts both efficiency and integrity risk. Maintenance decisions shape future flexibility.

Managing these interactions dynamically requires more than isolated tools.

Applied computing’s Orbital platform addresses this by combining physics-based process models, time-series operational intelligence and language-based reasoning within a single unified architecture.

Rather than deploying disconnected optimisers, Orbital enables intelligence to be layered brick by brick across an asset while remaining connected at the system level. Each additional capability strengthens the overall system rather than creating new silos.

From reactive management to predictive performance:

The implication of this approach is structural. Instead of layering digital tools onto existing workflows, system-level intelligence becomes the organising principle. Maintenance, integrity and operations become interconnected dimensions of performance rather than isolated functions.

For the first time in my career, I believe we have technology capable of identifying and capturing performance gaps while they are still recoverable.

This does not replace engineering expertise. It amplifies it.

Making the transition practical:

For organisations looking to accelerate this shift, several principles matter:

• Start with high-value constraints where cross-unit interaction drives margin.

• Build on shared data foundations rather than isolated pilots.

• Integrate physics and operational history to maintain trust.

• Deploy incrementally, expanding capability brick by brick.

• Ensure intelligence feeds directly into operational workflows, not separate dashboards.

The next decade will likely see AI evolve from experimental pilots to an integrated operating layer across energy assets. The winners will not simply deploy more algorithms. They will unify architecture, scale intelligence and align processes around system-level performance.

After 35 years in manufacturing, that feels less like incremental digital progress and more like a generational shift.

Getting ahead of hindsight is no longer aspirational.

It is becoming operational.