Water Utilities Know Where the Leak Is. Do They Know What It Will Break?

The technology keeps improving. Free-swimming inspection tools now traverse live water mains, detecting leaks and gas pockets without service interruption. Distributed acoustic sensing identifies pipe anomalies across kilometres of buried infrastructure from fixed sensor arrays. Machine learning models trained on decades of break records flag high-risk segments before they fail.

Digital investment in water utilities is accelerating to match: in the United States, investment in digital water solutions is projected to grow at 6.5% annually, reaching USD 10.8 billion by 2030, with asset management claiming roughly 18% of that spend.

And yet, finding an anomaly is only step one. The harder question — what will this anomaly cause if it goes unaddressed for eight weeks? — requires intelligence that most utilities do not yet have.

What the new sensors see, and what they don't

Consider a typical finding: a pressure drop reading on a distribution main, confirmed by in-pipe inspection as a corrosion patch roughly 40 metres long. A genuine detection that previously would have required a service interruption or educated guesswork.

Now the work order queue opens. Someone enters the anomaly into the asset management system. And here is where most utilities hit a wall.

The GIS-based linear asset management tool shows the segment, its inspection history, and adjacent segments. What it cannot show is that the main feeds a booster pump station whose duty pump has been running in alarm condition for eleven days, with the standby pump in scheduled maintenance until the end of the month. It cannot show that the pump station feeds a water treatment facility currently running at 91% of its design throughput due to seasonal demand. It cannot model the pressure consequence for the 14,000-property zone downstream if that corrosion patch progresses to a partial burst over the next six weeks.

That information lives in a separate CMMS. The two systems do not share a data model. The consequence analysis does not happen — or it happens in a spreadsheet, built by someone with deep institutional knowledge who currently does not have the time.

Why the two-system architecture fails at exactly this point

The architecture problem is structural, not incidental. Linear assets — pipes, cables, rail corridors, electric transmission lines — are defined by network topology. Risk propagates through connectivity. Analyzing a linear asset means reasoning about the graph: which segments are adjacent, what flows are at stake, how pressure distributes if a node is compromised. Managing linear assets well requires GIS-native models, dependency-graph traversal, and cascade failure simulation built into the core of the platform.

Discrete assets — pump stations, treatment facilities, compressor stations, substations — are defined by hierarchy. Performance is tracked through parent-child relationships: the facility contains systems, systems contain equipment, equipment contains components. Failure modes are classified by ISO 14224 taxonomy. OEE-grade telemetry flows up the hierarchy. Managing discrete assets well requires a different reasoning model — one that understands operational context in hierarchical terms.

Most water utilities have ended up with one system for each. A decade ago, this was a defensible technology decision: tools were specialized, and integration was assumed to be solvable through middleware. In practice, middleware delivers periodic data synchronization and duplicated work orders. It does not deliver real-time joint reasoning. When an anomaly is detected in the pipe network, the downstream consequence analysis still has to be assembled manually — because no single reasoning layer can see both the network and the facilities it feeds.

Artificial intelligence needs a complete picture

The EAM market is in a genuine AI transition. Machine learning models now forecast equipment failures with accuracy rates exceeding 85%, with some systems projecting failure windows six to twelve months out. That capability is valuable — but only if the model has access to the full operational context.

An AI model predicting maintenance needs for a pump station, fed only the pump station's data, will produce a prediction for the pump station. Feed it the pipe network's current anomaly register alongside that data, and the prediction changes: the maintenance window, the urgency score, the required crew profile — all of it shifts when the AI can see that the asset's supply network is already under stress.

UniEAM's agentic platform is designed to operate across both data models simultaneously. AI agents traverse network-aware models for the pipe system and hierarchy-aware models for the treatment facility in a single reasoning pass, producing recommendations that reflect the actual operational picture rather than one half of it. Every recommendation carries a full decision log — which matters increasingly as regulators expect utilities to demonstrate not just what was decided, but why.

The audit-ready standard is rising

CISA's CIRCIA framework and EPA water sector security guidance are both moving in the same direction: visible, auditable risk management across the full asset estate. Utilities investing in the latest in-pipe sensor technology while managing their treatment plants in a separate CMMS will struggle to produce the unified risk narrative that regulators expect. The detection investment is sound. The infrastructure around it — the EAM layer that contextualizes each finding, models the downstream consequence, and produces a defensible intervention plan — needs to span the whole portfolio.

For utilities managing both linear networks and discrete facilities, that means one platform with genuine reasoning on both sides. For a deeper look at the analytical foundations of network-asset modelling in water and utilities, see our guide to linear asset management.

---

Curious how UniEAM handles both the network and the plant? Request a demo.