The challenge
The utility's non-revenue water rate had reached 28% — above regulatory target and still rising — despite two previous waves of technology investment. Acoustic loggers and pressure-management upgrades had both been deployed, but without an integrating layer, each technology produced its own alert list with no reconciliation between them. Field crews were running two parallel investigation backlogs, and the combined hit rate on investigations was low. The question for leadership was not whether to invest further in technology, but how to make existing technology work more effectively.
Discovery
A district-level pilot in the worst-performing zone demonstrated the integration value within eight weeks. Leaks the acoustic loggers had flagged with low confidence became high-confidence findings when correlated with the pressure transient in the same district metered area. The reverse was equally informative — pressure transients without an acoustic signature were almost always pump events, not leaks, and could be deprioritised cleanly. The analytical case for fusion was established in eight weeks, with measured investigation hit rate as the proof.
Our approach
Fusion model design: built a per-DMA fusion model combining acoustic logger signal, pressure-transient telemetry, and consumption pattern to produce a single ranked leak-likelihood list
Investigation programme redesign: replaced two parallel field-investigation backlogs with one prioritised work queue, with the cost of each investigation shown against the expected water saving
GIS integration: connected the model output to the existing GIS so field crews work from the same map they already use and trust
Continuous improvement programme: established a monthly recalibration cycle against confirmed leak repairs so the model improves alongside the operator without requiring re-engagement
The Outcome
The engagement delivered a leak-intelligence layer that fuses acoustic, pressure, and consumption telemetry per district metered area, produces a single ranked investigation queue, and recalibrates monthly against confirmed repairs. The field organisation retained its existing tools — investigation routing, GIS, work order — and gained a single, reliable starting point for what to investigate next.
The results
“The win was not a new sensor. The win was making the sensors we already owned work together. That is what changed our crew's hit rate.”
