A water utility never sleeps. The utility runs treatment and distribution infrastructure that has to perform every hour of every day, across facilities spread over hundreds of miles, much of it built decades ago. Keeping it running meant sending experienced engineers to drive between plants, clipboard in hand, to inspect equipment that mostly looked fine right up until it didn't.

The senior people who could read a pump by its sound and vibration were the same people stuck in a truck for three hours a day. Knowledge was trapped in travel, inspections were periodic rather than continuous, and the first sign of many failures was the failure itself.

The challenge

The utility came to us with a deceptively simple question. Could an expert inspect a plant without being in the plant, and could the plant tell us when something was about to break? The constraints were not simple at all. Equipment from five decades and a dozen vendors. Facilities with patchy connectivity. A safety-critical, regulated environment where "move fast and break things" is exactly the wrong instinct. And a workforce that, rightly, would not trust a black box.

The approach

We didn't start with a headset. We started in the field, walking real inspection routes with real technicians to learn what an expert actually looks at, listens for, and writes down. Only then did we design AquaSight AR, a platform that captures that expertise, makes it remote, and puts a predictive layer underneath it.

Architecture: field AR walkthrough feeds telemetry capture, which feeds predictive failure models, which feed a remote command center
FIG.02Four layers, field walkthrough, telemetry capture, failure prediction, and a remote command center, wired into one continuous loop.
01
A walkable digital twin
Every facility was spatially mapped so an engineer at HQ can move through it in AR. On site, a technician sees the same overlays anchored to the real equipment in front of them through a headset or iPad.
02
Live telemetry on every asset
Vibration, pressure, temperature, flow, and runtime stream from IoT sensors into a unified model of asset health, turning a once-a-quarter inspection into a continuous one.
03
Failure prediction, not just dashboards
Models trained on historical maintenance and sensor signatures flag the early fingerprints of bearing wear, cavitation, and seal failure, each with a confidence score and an estimated time to failure.
04
A remote command center
One operations view across all nine facilities, so the most experienced engineers triage the whole fleet, route crews to what actually needs them, and log every inspection back to the system of record.

The win wasn't the headset. It was turning scarce expertise into something that scales across every plant, every shift, at once.

Crucially, we built for trust. Every prediction shows its evidence, the sensor trend behind the call, so a technician can agree, disagree, and teach the system. Adoption was designed in from week one: training, side-by-side rollout, and a deliberate choice to augment the expert, never to replace their judgment.

First-person AR inspection view of a valve, with a bounding box, a 92 percent health ring, nominal status, and a 30-day vibration trend
FIG.03The technician's view: any asset, framed and identified, with a live health score, status, and history, plus one tap to log the inspection.

The outcome

Within the first year in production, AquaSight AR changed the economics of inspection. Unplanned outages fell as failures were caught in their early, cheap-to-fix stage, a median of eleven days before they would have taken a device offline. Routine drive-time collapsed as engineers inspected remotely and dispatched crews only when the data warranted it.

The plant stopped being a place you had to drive to. It became something you could read, anywhere, before it broke.

More quietly, the platform began to preserve something the utility had been losing: the instinct of its most experienced people, captured as data and shared with every crew that follows. That is the part no spreadsheet on day one could price, and the part American Water now runs on every day.