UK water companies lose roughly a quarter of the water they treat before it ever reaches a tap. Leakage has been an Ofwat priority for years, and the regulatory pressure intensified sharply through the PR24 price review, which tied water company revenues to leakage reduction targets running to 2030. The technology being deployed to address this is a useful case study in what IoT and edge computing look like when they’re solving a genuinely large-scale infrastructure problem.
The Scale of the Problem
Water distribution networks are enormous, old, and almost entirely underground. The UK has around 340,000 kilometres of water mains, most of which have no real-time monitoring. A pipe can leak for weeks or months before a burst large enough to appear at the surface — at which point a significant volume of treated, pumped, and pressurised water has already been lost. Thames Water’s leakage figures have attracted the most public attention, but all of the major water companies are dealing with ageing infrastructure at scale.
Detecting leaks earlier requires sensors distributed across the network. The challenge is that distribution networks run through places where power and connectivity are patchy: rural roads, beneath city pavements, across open farmland. Sensing technology has to be low-power enough to run on batteries for years, reliable without maintenance, and capable of transmitting data over whatever network happens to be available.
The Sensing Layer
Two types of sensors form the backbone of smart water leak detection:
Acoustic correlators attach to accessible pipe points — hydrants, valves, meter chambers — and listen for the characteristic sound signature of water escaping through a crack or joint failure. Modern systems compare signals from multiple correlators simultaneously, allowing algorithms to triangulate leak location to within a few metres. The correlators are battery-powered and transmit via NB-IoT or LoRaWAN, both of which give multi-year battery life at the data rates needed for sensor telemetry.
Pressure transient loggers monitor pressure fluctuations in the network continuously. A sudden pressure drop indicates a possible burst; sustained low pressure in a defined zone suggests ongoing background leakage. Placed at intervals across a district metering area — a subdivided section of the network with flow measurement at each boundary — they build a picture of network health that network operations teams can monitor remotely.
Where Edge Processing Changes the Equation
Raw sensor data from an acoustic correlator is voluminous. The audio frequency data needed to characterise a leak signature would fill a network connection quickly if transmitted raw, and would drain batteries in days rather than years.
Edge processing changes this. A correlator with on-device signal processing analyses acoustic data locally, identifies the frequency patterns associated with leakage, and transmits only a structured result — “possible leak at grid reference X, confidence 80%, estimated flow rate Y litres per hour” — rather than the raw waveform. The central platform receives structured alerts instead of raw data, battery life extends to years, and the cellular data cost drops to almost nothing.
This is TinyML applied to a practical problem: a small machine learning model trained on leak acoustic signatures, running on an embedded processor inside a device attached to a valve chamber in the ground, classifying incoming sensor data in real time without a network connection. The classification happens on the device; the alert travels over NB-IoT when connectivity is available.
UK Deployments
Severn Trent has been among the more public about its smart network investments. Its acoustic sensing programme has deployed sensors across its Midlands network as part of a district metering strategy that subdivides the network into smaller pressure zones. Earlier detection translates directly to less water lost per incident and lower repair costs from secondary pipe damage.
Anglian Water operates in a region where water scarcity is a more acute concern than in much of the UK. Its smart metering programme for domestic properties — one of the most advanced in the sector — allows it to identify leakage at the individual property connection level, which accounts for a meaningful share of total network losses. Combining household smart meters with network monitoring creates a picture detailed enough to prioritise repair work geographically.
Thames Water, despite its financial difficulties, has committed to significant smart metering and network monitoring investment in its PR24 business plan, partly because Ofwat’s final determination made leakage reduction targets explicit conditions of its price settlement.
The Ofwat Driver
Ofwat’s PR24 final determination set leakage reduction targets through to 2030, with financial penalties for underperformance and financial rewards for beating targets. Smart network monitoring is the primary mechanism through which water companies are planning to meet those targets — the alternative, reactive repair of leaks that have already surfaced, is simply too slow to achieve the required reductions at the required scale.
The regulatory pressure has directly driven capital allocation. Most large water companies’ 2025–2030 business plans include substantial allocations for network sensing infrastructure, smart metering rollout, and the data platforms needed to act on sensor data operationally.
What This Looks Like for the IoT Sector
Water network monitoring represents a significant deployment at scale for several technology categories: NB-IoT and LoRaWAN network operators, battery-powered sensor manufacturers, edge AI chip vendors, and IoT platform companies. The technical requirements are demanding — sensors must operate reliably for five or more years without physical access, in underground environments, across connectivity conditions that range from excellent urban NB-IoT coverage to rural locations where LoRaWAN is the only option.
The deployments that are working demonstrate that the technology has reached the maturity the problem requires. That’s a useful signal for anyone thinking about where the next large-scale IoT infrastructure investment is going.