Water Utilities Are Running Two AI Programs. They Need One.

Water is the infrastructure sector getting the most attention from AI vendors right now — and with good reason. The global AI in water and sanitation market is projected to reach USD 24.45 billion by 2031, growing at 26.8% CAGR. US utilities alone are increasing digital investment at 6.5% annually. Across the sector, operators adopting AI-driven leak detection and pump optimisation are reporting energy reductions of 15–25% and measurable extension of asset life.

The momentum is real. But it is hiding a structural problem that will limit the returns most utilities actually see.

Water utilities operate two fundamentally different categories of assets. The first is the distribution network: hundreds or thousands of kilometres of buried mains, pressure mains, service laterals, valves, and hydrants that move treated water from source to tap. These are linear assets. Their behaviour is governed by topology — pressure zones, flow dynamics, network connectivity. A failed main doesn't just stop serving its segment; it changes flow patterns upstream and downstream, creates pressure transients, and may shift hydraulic loads in ways that stress adjacent mains and the pump stations that feed them. Managing linear assets well requires GIS-native data models, network-graph reasoning, and the ability to simulate how a condition event at one node propagates through the connected system.

The second category is treatment and pumping infrastructure: water treatment plants, pump stations, storage tanks, reservoirs, and ancillary mechanical and electrical plant. These are discrete assets. They are hierarchical — a facility contains systems, systems contain equipment, equipment contains components — and their failure modes follow parent-child patterns. Discrete asset management is the domain of ISO 14224 taxonomies, OEE-grade telemetry, and maintenance optimisation that accounts for equipment interdependency within a process train.

Both categories are central to what a water utility does. Both need AI. And in most utilities today, both have it — in separate systems that were never designed to talk to each other.

The buried asset problem

Over 70% of a water utility's assets — its pipe network — are buried. You cannot see them. You cannot touch them without excavation. What you can do is infer their condition from operational signals: pressure differentials, flow imbalances, acoustic signatures, and historical patterns of breaks and repairs on similar mains in similar soil conditions.

That inference requires all available signal. A significant share of that signal comes from the discrete assets connected to the network: pump station performance, treatment plant throughput, pressure zone controller readings. If your pipe network AI cannot see pump station operational data — and it cannot, if they live in separate systems — then your condition models for buried assets are systematically underinformed.

The dynamic runs the other way too. Pump cavitation, a common and damaging failure mode in water distribution, is partly a function of hydraulic conditions in the network feeding the pump inlet. A pump running in a cavitation-prone regime may be doing so because of a pressure deficit caused by a partially occluded main three kilometres upstream. A discrete asset maintenance system with no visibility into the pipe network will generate a pump maintenance recommendation without ever identifying the actual cause. The pump gets serviced. The main gets worse. The pump cavitates again.

Neither AI was wrong. Both were just half-informed.

Where the AI investment goes

Industry data shows that US digital water investment breaks down approximately as: 41% on metering, 20% on network management, and 18% on asset management. The network management bucket funds GIS-adjacent linear tools — flow modelling, hydraulic simulation, SCADA integration for the distribution network. The asset management bucket funds CMMS-style platforms for the treatment and pumping facilities.

Two systems. Two AI projects. Two vendor relationships. And two data models that were never designed to share an asset register or a reasoning layer.

When each AI sees only half the operational picture, the gap between them doesn't stay clean. It becomes a persistent blind spot: failures that should have been preventable, recommendations that seem locally correct but are contextually wrong, and compliance reports that require manual reconciliation before they can be filed. As CISA's evolving OT asset inventory guidance makes clear, regulators are no longer satisfied with a utility's ability to account for either its network assets or its facility assets. They want visibility across all of it.

What unified looks like in practice

The UniEAM agentic platform models both linear and discrete assets in a single, unified asset graph. Linear assets — mains, pressure zones, network segments — are modelled with full GIS-native topology. Discrete assets — treatment plants, pump stations, process equipment — are modelled with hierarchy-aware taxonomy aligned to ISO 14224.

AI agents running across this unified graph can form questions that are simply not possible in a split-system environment. When pump performance degrades, the agent immediately correlates with network pressure data to determine whether the root cause is in the pump itself or upstream in the distribution network. When a main shows early-stage pressure anomalies, the agent assesses the hydraulic stress implications for the pump stations serving that zone and adjusts their inspection priority accordingly. The cascade logic runs across the boundary that most EAM platforms treat as a wall.

Every inference carries a full decision log: which data was used, which model generated the recommendation, and what the confidence bounds were. This isn't a separate audit report — it is an intrinsic output of how the platform reasons. For utilities navigating increasingly demanding OT asset inventory requirements, that auditability matters.

The water sector's digital investment is substantial and growing. Whether it compounds into genuine predictive intelligence or plateaus into two parallel toolsets that can't see each other depends on where the asset graph lives. One place, or two.

Learn more about how UniEAM approaches linear asset management in water networks, including network-aware condition assessment for buried pipeline infrastructure.

Curious how UniEAM handles both? Request a demo.