Somewhere in your network right now, a section of trunk main has made it onto next year's replacement list for exactly one reason: it is old enough. Not leaking. Not flagged by a single sensor. Not sitting in the fault log. Just old enough, and that was sufficient. Dig it up anyway, and there is a fair chance you will find sound iron, years of genuine service life still in it, and a full replacement invoice where a repair job would have done. That is not a technical judgement

Image source: Metro UK, "Burst water pipe in Finsbury Park turned roads into rivers" (2019) Most bursts are not surprises. They are the final event in a deterioration sequence that began long before water reached the surface. That sequence generates detectable signals: and utilities that know how to listen and interpret them can avoid most pipe breaks. That is the core argument behind leak before break: the gap between first signal and catastrophic failure is not an inevitab

For water utilities, AI is not about chatbots or image generation. It is about interpreting sensor signals from complex, buried infrastructure. This article explores how domain-specific machine learning trained on real network data can help detect emerging leaks, reduce false alarms and support earlier operational decisions.

Adelitics exists because we’re not comfortable with “trust us” technology. For more than a decade, our team has been developing and validating domain-specific machine learning and signal analytics for water networks, not just using generic AI tools.