Predictive Maintenance Is Now the Standard. So Why Are Mixed-Asset Operators Still Flying Blind?

Predictive maintenance is no longer a competitive advantage — it is table stakes. An estimated 88% of manufacturers are now applying predictive analytics to their assets, using machine learning to identify degradation patterns before failures occur. The techniques work: AI models can forecast asset failures six to twelve months in advance with accuracy rates above 85%.

And yet, for a significant category of asset-intensive operators, the technology is delivering less than half of what it should. Not because the models are wrong, but because they are working with half the data.

The portfolio gap

Utilities, transit authorities, and energy companies don't manage a single asset type. They manage portfolios that span two fundamentally different categories: linear assets like pipelines, transmission lines, water distribution networks, and rail corridors — alongside discrete assets like pumping stations, substations, manufacturing cells, and maintenance depots.

Each category obeys different physics, fails in different patterns, and requires a different data model to reason about effectively.

Linear assets behave as networks. A pressure drop in a water main propagates through the distribution topology, affecting pump stations, pressure zones, and service areas downstream. Understanding the risk of any single anomaly requires reasoning across the graph — which segments are adjacent, which facilities depend on that section, how a failure would cascade. GIS-native modeling, dependency graphs, and cascade failure simulation are not optional features for these assets; they are the baseline requirement for a coherent risk assessment.

Discrete assets behave as hierarchies. A facility contains systems; systems contain equipment; equipment contains components. Failure modes follow parent-child relationships. A bearing failure in a pump matters differently depending on the criticality of that pump within its system, and the system within the facility. Effective reasoning here runs up and down the hierarchy — ISO 14224 taxonomies, OEE-grade telemetry, and parent-child maintenance correlation.

Most predictive maintenance deployments handle one of these categories well. The problem is that the operator's portfolio contains both.

What the boundary looks like in practice

Consider a water utility managing a regional distribution network alongside several treatment plants. The network is linear: pipes, mains, pump stations connected across geography. The treatment plants are discrete: process equipment, chemical dosing systems, and filtration units in a hierarchical facility structure.

The utility has probably deployed predictive analytics for pipeline condition — pressure monitoring, acoustic leak detection, corrosion modelling. They may also have predictive models running on treatment plant equipment. Both sets of models are generating useful signals.

What neither can do, in isolation, is correlate those signals. When a pipeline section shows early degradation, the operationally relevant question isn't just "when will this pipe fail?" It's: which treatment plant does this pipeline feed? What is the current maintenance state of that plant's intake infrastructure? If both the pipe and the intake pump are at elevated risk simultaneously, the failure scenario and the maintenance priority are different from either considered alone.

Answering that question requires both data models to be live in the same reasoning environment — not synchronised nightly via ETL, not reconciled quarterly in a spreadsheet, but genuinely unified at the data layer so an AI agent can traverse the network graph to the asset hierarchy and back in a single pass.

The 25–35% that goes uncaptured

Operators who have moved from separate linear and discrete EAM systems to unified platforms consistently report operational efficiency gains in the range of 25 to 35 percent. That delta is not primarily from eliminating duplicate licence costs, though that matters. It comes from improvements in decision quality when AI can reason over complete portfolio data.

NextEra Energy's deployment of AI across its generation and transmission assets — integrating proprietary asset data with machine learning — shifted maintenance scheduling from reactive to predictive, factoring in equipment condition, crew availability, supply chain constraints, and weather exposure simultaneously. That kind of multi-dimensional reasoning is only possible when the data it draws from is structurally unified, not loosely connected.

Digital investment in water infrastructure alone is projected to reach $10.8 billion by 2030, growing at 6.5% annually. The EAM software market overall is on track to reach $19.68 billion by 2030 at a 17.2% annual growth rate. A significant share of that capital will be spent solving exactly this problem: retiring the two-system architecture in favour of platforms capable of handling both asset topologies, or paying to integrate systems that were never designed to share a data model.

The latter path is common. It is also structurally limited.

Why native beats integrated

The standard response to the portfolio gap is integration: connect the linear system to the discrete system via APIs, align the asset IDs, run a shared reporting layer on top. This is achievable, and some organisations have done it well. But it carries a ceiling.

Integration connects two data models. It does not unify them. An AI agent reasoning across an integrated pair of systems has to manage the impedance between two schemas, two ID spaces, and two sets of semantic assumptions every time it crosses the boundary. That overhead accumulates — particularly when the scenario being evaluated involves assets on both sides of that boundary simultaneously.

The UniEAM agentic platform holds both data models natively: the network graph for linear infrastructure and the asset hierarchy for discrete facilities in a single environment. AI agents working within that environment traverse both without translation overhead, and every recommendation they produce is logged with its complete decision path — which data inputs it drew from, which model version generated the output, and what risk threshold triggered the action.

That decision log is not an afterthought. In an environment where FERC, CISA, and the March 2026 SOCI Act amendments are all pushing toward mandatory, auditable OT asset management, a prediction your compliance team cannot explain is a prediction you cannot act on with confidence.

For a deeper look at how network-aware and hierarchy-aware reasoning differ in practice, see our linear asset management knowledge guide.

The question worth asking

If your organisation manages both pipeline infrastructure and plant or facility assets, the question is not "do we have predictive maintenance?" Almost certainly, you do. The question is: "Can our AI reason over both sides of the portfolio in the same operation?"

If the answer is no, you have found the ceiling.

Curious how UniEAM handles both? Request a demo.