The State of Things
A global technology company operating a geo-mapping fleet across multiple continents. Thousands of sensor-equipped vehicles collecting street-level imagery and spatial data. Each vehicle carried a sensor array generating continuous telemetry — GPS accuracy, camera health, LiDAR calibration, storage capacity, network connectivity. When a sensor degraded, the data it collected degraded with it. But the vehicle kept driving.
Sensor faults were discovered in post-processing — hours or days after collection. By then, the route had been completed with compromised data. The recollection cost wasn’t just the fuel and driver time. It was the scheduling complexity of re-routing a vehicle through the same coverage area, often in a different season or lighting condition. Some routes couldn’t be recollected at all within the planning window.
The Mission
“Every sensor fault detected in real time. Every degraded collection flagged before the route is complete. Maintenance actions staged and approved before the vehicle returns to depot.”
What We Found
Fourteen percent of collection routes contained segments where at least one sensor had operated outside specification. The fault signatures were present in the telemetry — but the telemetry was monitored in batch, not in real time. A camera calibration drift that developed at mile forty was not detected until the post-processing pipeline flagged the imagery at mile two hundred.
Maintenance scheduling was reactive. Vehicles returned to depot on a fixed rotation. Sensor issues discovered in post-processing were queued for the next scheduled maintenance window — which might be days away. During those days, the vehicle continued collecting data with a known fault. The maintenance team knew this. The operations team knew this. The data didn’t lie. But the workflow did nothing with the data in time.
The Personas
Coordinates collection routes across regions and seasons. Knows which vehicles have issues — eventually. Needs to know before the route is wasted, not after.
Maintains the fleet’s sensor arrays on a fixed schedule that doesn’t account for real-time fault data. Every reactive repair is a missed proactive one.
Owns the collection quality standard. Catches degradation in post-processing and orders recollection — a decision that would have been unnecessary if the fault had been caught in the field.
The Build
The governance model defined sensor health thresholds, fault classification rules, and escalation policies. What constituted a critical fault versus a degraded-but-acceptable reading. Which faults required immediate route termination. Which allowed continued collection with a quality annotation. The rules were codified from the maintenance team’s institutional knowledge — the same decisions they made manually, now enforced systematically.
The detection layer ingested sensor telemetry in real time, comparing readings against the governance thresholds. When a fault was detected, it classified the severity, calculated the data quality impact on the current route, and determined the appropriate response. For critical faults: immediate alert with route modification recommendation. For degraded sensors: continued collection with quality annotation and maintenance action staged for depot return.
The Portal
A fleet command view. Every vehicle’s sensor health visible in real time. Fault alerts arriving with their classification, their data impact assessment, and their recommended action — pre-staged for approval. The operations manager didn’t diagnose the problem. She reviewed the recommendation and approved the response.
Maintenance actions were queued automatically. When a vehicle returned to depot, the maintenance team already had the work order: which sensors, which faults, which parts. The diagnostic had happened in the field. The depot visit was for the repair, not the investigation.
The Signal
Fourteen percent of collection routes had been operating with undetected sensor faults. Real-time detection eliminated the gap between fault occurrence and fault response. Maintenance actions pre-staged before depot return. Recollection costs avoided by catching degradation before the route was complete.
What This Opened
The fault data, collected continuously across the fleet, revealed patterns that fixed-schedule maintenance never could: sensor degradation curves by vehicle age, environmental conditions that accelerated specific fault types, route characteristics that stressed particular sensor configurations. The next step was predictive — scheduling maintenance not when a sensor failed, but when the data said it was about to.
The Engagement Arc
Fourteen percent of routes had been operating with undetected sensor faults. The engagement closed the gap between fault occurrence and fault response — and the next move is to predict the failure before it happens.