A quiet street, an unseen underground network, cubic metres disappearing without leaving a puddle. For a water utility, a leak often resembles an investigation with no witnesses. The digital twin promises to shorten the search: represent the network, monitor its behaviour and flag problems before sending in an excavator. Looking ahead to September 2026, however, the challenge is not to make pipes look impressive on screen. It is to determine whether this additional intelligence saves more water than well-managed conventional monitoring equipment.
A network losing water—and information
In France, data from the Observatory of Public Water and Sanitation Services have long documented losses amounting, nationally, to roughly one litre in five entering the distribution network. This average conceals very different situations. A dense urban network under constant monitoring faces neither the same constraints nor the same repair costs as a rural utility whose pipes stretch for kilometres to serve only a few customers.
The Water Plan unveiled in 2023 put leak reduction back among public policy priorities. But replacing pipes requires time, money and knowledge of infrastructure assets that is sometimes incomplete. Old maps may incorrectly locate a pipe; a valve assumed to be open may be closed. Even before predicting a burst, operators must therefore reconcile the actual network with the one they believe they have.
The digital twin: more than an animated map
A geographic information system shows where pipes run. A hydraulic model calculates how water flows through them, based on pipe diameters, consumption, reservoirs, pumps and valves. The digital twin adds a feedback loop that regularly checks the model against field measurements. Its fidelity depends less on a three-dimensional display than on the quality of these updates.
Hydraulic simulation tools are not new. EPANET, developed by the US Environmental Protection Agency, has been used since the 1990s to model distribution networks. The innovation lies mainly in integration: more accessible telemetry, communications, computing power and software that bring together measurements, simulations and maintenance work. A hydraulic model does not become reliable simply because it is labelled “artificial intelligence”.
What sensors really tell us
Flow meters measure the water entering a district. Pressure sensors reveal unusual variations. Acoustic loggers look for the sound signatures of leaks. Smart meters can help distinguish an increase in consumption from a loss in the network, provided the appropriate data are available. No instrument sees everything: an isolated measurement is often ambiguous.
Imagine a district where night-time flow increases when demand is normally low. The system compares this increase with observed pressures and the model’s forecasts. It can then suggest a search area. But overnight industrial activity, a reservoir being filled or a faulty sensor can produce a similar anomaly. Technicians must be able to understand why an alert has appeared, rather than simply receiving a red warning light.
Locating leaks before digging, without promising infallibility
The operational promise is simple: reduce the area requiring inspection. Instead of covering an entire neighbourhood with listening equipment, the team starts with a few priority pipe sections. Simulations can also help prepare the work: which valves should be closed, how many customers will be affected, and what pressure will remain available elsewhere? The potential benefit goes beyond leak detection alone.
Three tasks must nevertheless be distinguished. Detecting an anomaly means identifying that the network is behaving differently from expectations. Locating a leak means reducing geographical uncertainty. Predicting a burst means assessing a future risk. This last task draws particularly on age, material, past incidents, soil conditions and pressure variations. It remains probabilistic: a section deemed vulnerable may last a long time, while another may burst without any actionable warning.
The real test: how much water is saved?
A successful demonstration is not enough to establish a business case. Finding more leaks may simply reflect a more intensive search campaign. And a fall in the volume of water distributed may result from restrictions or a less dry summer. To attribute a gain to the digital twin, the local authority must establish a baseline and account for changes in consumption, pressure and the area covered.
- Measure physical outcomes: estimated volumes lost, time between a leak occurring and being repaired, and changes in night-time flows.
- Track operational efficiency: confirmed alerts, unnecessary site visits, time taken to locate leaks and callouts avoided.
- Calculate the full cost: sensors, communications, software, integration, training, maintenance and model recalibration.
- Compare against a credible alternative: improved district metering, acoustic leak detection or targeted pipe replacement.
Water saved should not automatically be valued at the tariff charged to customers. For the utility, the immediate saving primarily consists of avoided abstraction, treatment and pumping costs. Depending on the context, this may also preserve water resources, free up production capacity or defer investment. These benefits matter, but they must be distinguished to avoid artificially inflating returns.
There are obstacles in the office, too
The first challenge is often organisational. Asset management, operations, IT and construction teams each hold part of the information. If a repair is not followed by an update recording a pipe’s actual diameter or a valve’s position, the model deteriorates. A useful digital twin therefore demands ongoing discipline, with clear responsibilities and time devoted to data quality.
Public procurement must also anticipate the end of the contract: who owns the data, in what format can they be retrieved, and can the model be retained when switching providers? Cybersecurity becomes essential once the platform communicates with operational systems. A diagnostic tool does not necessarily need to control pumps. Separating functions and restricting access reduces risks.
Start small, demonstrate results, then expand
For a local authority, a sensible approach is to choose a well-documented district with an identifiable problem and a team capable of taking action. The pilot must cover a variety of operating conditions, not just a few favourable weeks. Above all, repairs must be funded: faster detection saves water only if teams can act. Digital technology guides the work; it replaces neither pipes nor staff.
What next? Looking ahead to September 2026, the most useful progress could be less spectacular than an autonomous network: better-maintained models, explainable alerts and public procurement tied to verifiable results. Local authorities would do well to ask not how much data a platform takes in, but how many days of leakage it can prevent. The best digital twin will be the one that helps teams dig in the right place, at the right time—and whose usefulness can be proven.
