The AI EAM Gap That New Regulations Just Made Impossible to Hide

Last month, NERC CIP-003-11 took effect. The new Cyber Security—Security Management Controls standard extends mandatory compliance obligations to low-impact BES Cyber Systems — substations, remote facilities, and field assets that previously sat outside the tightest governance requirements. CISA followed with updated OT asset inventory guidance that doesn't distinguish between your transmission network and your control room: it wants an auditable, unified register of every operational technology asset across the organisation.

For a utility managing both a water distribution network and a treatment plant, or an energy company operating pipelines alongside compressor stations, this is not an abstract compliance exercise. It's a practical test of whether your EAM architecture can produce a single, coherent picture of your asset estate — or whether it will reveal that you've been running two systems, with two data models, that have never been designed to talk to each other.

Most will fail that test. Not because the data doesn't exist, but because it lives in the wrong shape, in the wrong system.

Two asset types. Two mental models. Two incompatible systems.

The deepest structural problem in utility EAM is that linear assets and discrete assets don't behave the same way — and most platforms were built to handle one or the other, not both.

Linear assets — pipelines, transmission lines, rail corridors, water distribution mains — are spatial and topological. Their failure modes propagate. A degraded section of pipe exposes the pressure zones, pump stations, and customer branches connected downstream. Managing them well means GIS-native data models, network graph reasoning, and the ability to simulate cascade failures before they happen. The right question when a sensor flags an anomaly is not just "what will fail?" but "what else fails when it does, and in what order?"

Discrete assets — treatment plants, substations, compressor stations, maintenance depots — are hierarchical. A facility contains systems, systems contain equipment, equipment contains components. Failures follow parent-child relationships. Performance is benchmarked against ISO 14224 taxonomies. The telemetry stack is designed to aggregate OEE-grade signals up through a structured hierarchy, not a network graph.

Both models are correct. Both are necessary. The problem is that in most mixed-portfolio organisations, they're managed by two separate platforms with separate data models, separate asset IDs, separate maintenance histories — and no shared reasoning layer.

The AI gap no one talks about

The business case for AI in asset management is compelling. Machine learning models can now forecast equipment failures six to twelve months in advance with accuracy rates above 85%. The global AI-driven predictive maintenance market, valued at $1.77 billion in 2025, is projected to reach $19.27 billion by 2032. More than two-thirds of maintenance teams plan to adopt AI before the end of 2026.

But a predictive model is only as useful as the data graph it can reason over. A model trained on pipeline sensor data cannot incorporate the maintenance state of the pump stations at either end of the network if that data lives in a separate system with no shared asset ID. A failure prediction for a treatment plant can't account for upstream network pressure dynamics if the distribution model lives in a different platform.

Fragment the asset graph, and you fragment the intelligence. The AI sees half the picture — and acts on it.

This is the gap that new regulations have made impossible to paper over. A CISA OT inventory audit or a post-incident NERC review doesn't accept "we have two systems and they don't sync." It requires a unified, traceable, auditable record of asset state, maintenance decisions, and the reasoning behind them. Producing that from two siloed platforms is a quarterly spreadsheet exercise — fragile, slow, and indefensible when someone asks why a particular inspection was deferred.

What unified intelligence actually enables

An EAM platform built to handle both asset topologies natively — not through middleware, not via nightly ETL jobs, but at the data model level — can answer questions that no siloed pair of systems can.

When a pressure anomaly surfaces in a distribution network, the system can simultaneously surface the maintenance history of connected pump stations, check whether scheduled inspections are overdue, model downstream exposure across the network topology, and generate a prioritised action recommendation. Not a chatbot response to a typed query. A structured agent action, proposed from a reasoning layer that holds the full asset graph, with a complete decision log attached.

That audit trail is not a compliance feature bolted on afterward. It's the natural output of a system where every prediction carries its data inputs, model version, and decision path as first-class metadata. When a regulator asks why a particular substation was flagged for priority maintenance in February, the answer is in the system — not in someone's memory, and not in a spreadsheet that's already been overwritten.

For a deeper look at how network-topology reasoning differs from hierarchy-based analysis — and why both are necessary for AI to work at scale — see our linear asset management knowledge guide.

The consolidation pressure is real

The EAM software market is expected to reach $8.22 billion in 2026 and grow to $17.46 billion by 2035. Analyst research consistently identifies unified platforms as delivering 3.2 times higher success rates compared to fragmented, point-solution architectures. The consolidation is already underway — driven not by technology enthusiasm, but by regulatory deadlines and the realisation that fragmented data is a direct compliance liability.

For operators managing linear infrastructure alongside discrete facilities, the two-system architecture is running out of runway. CIP-003-11 is live. CISA inventory requirements are not optional. The AI ROI that the market is pricing in depends on reasoning across the full asset estate — not half of it.

UniEAM's agentic platform was built from the ground up for this: network-aware reasoning for linear assets, hierarchy-aware reasoning for discrete assets, and an agent layer that operates across both — with a decision log on every action, by default.

Curious how UniEAM handles both? Request a demo.