AI in Water Utilities: Leakage, Assets and Regulated Service

Water utilities operate networks they cannot see, under public regulatory targets, with capital programmes that never cover everything needing replacement. The central problems — where is the water going, which assets are about to fail, how do we serve customers well — are all inference problems over data the utility already collects. This guide covers where AI helps, drawing on FISTA Solutions' AI agents work in utilities. It complements the energy and utilities whitepaper and how to build a maintenance work order agent. This article is general guidance, not regulatory or engineering advice.

Why is leakage the central problem?

Because it is simultaneously a regulatory target, a direct cost, and an environmental pressure. Water lost after treatment carries the full cost of treating and pumping it, abstraction is constrained, and performance is published.

Reducing it is the largest operational lever most water utilities hold, which is why detection and localisation attract disproportionate attention and investment.

Activity	Automatable	Qualified operator required
District flow anomaly detection	Yes	—
Leak localisation narrowing	Yes	Field confirmation
Asset condition inference	Yes	Engineering review
Renewal prioritisation	Yes, as recommendation	Programme decisions
Water quality determination	No	Yes
Supply interventions	No	Yes

What is actually hard about leakage?

Localisation. Night flow analysis at district level reliably indicates that a district is losing water, and that is detection. Finding the leak within a district of several kilometres of main means acoustic logging, correlation, and ultimately excavation.

Every metre of search narrowed before digging is money saved, because excavation in a road is the expensive part. Combining flow patterns, acoustic data, pressure signals, and burst history to prioritise where to survey is where the value concentrates.

Why must asset condition be inferred?

Because the assets are buried. Direct inspection requires excavation, which costs more than the information is usually worth, so condition has to be estimated from what is observable: installation date, material, soil chemistry, pressure regime, traffic loading, and the history of bursts nearby.

That is multi-factor inference over data the utility holds in fragments across asset registers, GIS, and maintenance records. Bringing it together is the substantive work.

What does renewal prioritisation achieve?

Better use of a constrained capital programme. Replacing pipe by age alone — the common default — renews sound assets and leaves deteriorating ones in the ground, because age correlates imperfectly with condition.

Prioritising by likelihood of consequential failure, weighted by what fails downstream, directs the same budget at the assets that matter. That is a planning improvement rather than an efficiency one, and it compounds over a capital cycle.

What about customer service?

High volume and concentrated around incidents. Supply interruptions, discolouration, pressure complaints, billing queries, and meter issues arrive constantly, and spike when something goes wrong in a network area.

Automating routine queries and, more valuably, proactively informing affected customers during an incident changes the experience of exactly the events that generate complaints. Customers accept interruptions they were told about far better than ones they discover.

What stays with qualified operators?

Water quality determinations, supply decisions, and network interventions. These carry statutory duty and public health consequences, and they require qualified judgement.

Systems should present better information to those operators — faster anomaly detection, assembled context, clearer asset history — rather than encroaching on the determination.

Who should own it?

Asset management and network operations jointly. Leakage sits with operations, renewal sits with asset planning, and the data that serves both is the same data. Separate ownership produces two systems built on the same sources with different definitions.

How is it evaluated?

Leakage reduction per intervention, excavation hit rate, asset failures on prioritised versus non-prioritised mains, customer contacts during incidents, and regulatory performance measures. Alerts generated is a metric that improves while nothing changes in the ground.

What goes wrong?

Detection without localisation, generating district-level alerts nobody can act on. Condition inference built on an asset register nobody trusts. Renewal prioritisation that engineering will not accept because the reasoning is opaque. And customer communication that starts after the complaints.

What does it cost to run?

Modest relative to the operational budget; the analysis is scheduled and the data volumes are manageable. The investment is in reconciling asset, GIS, and maintenance records into a usable model, which is data work and which most utilities need regardless.

What should you do first?

Check your excavation hit rate — the proportion of digs that find the leak where expected. That number measures how good your localisation currently is, and improving it has an immediate, quantifiable return.

How does metering change the picture?

Substantially where smart metering exists. Consumption data at property level makes district analysis far more precise, distinguishes genuine leakage from unrecorded consumption, and enables customer-side leak alerts that reduce both loss and bills.

Where metering coverage is partial, the analysis must handle mixed visibility honestly rather than extrapolating metered behaviour across unmetered properties, which is a common source of confidently wrong district estimates.

How FISTA Solutions helps

FISTA Solutions builds water utility systems that combine flow, acoustic, pressure, and burst history to narrow leak localisation, infer asset condition from reconciled registers for renewal prioritisation, and inform customers proactively during incidents, while quality and supply decisions stay with qualified operators, through AI agents, AI enablement, and forward deployed engineers. The record behind the approach is 150+ projects for 50+ companies with 47% efficiency gains.

To find leaks with fewer holes in the road, message FISTA on WhatsApp, or read the energy and utilities whitepaper.