District metering is the industry's answer to leakage. About half of Portsmouth Water's network does not have it, and is not going to get it. So the useful question is not whether district metering works. It is whether a burst can be found and located without doing it first.
Portsmouth Water set the test. We ran it on the monitors already in their ground. Below is what came back, and where it does not work.
First presented at the Global Leakage Summit, London, 3 September 2026, by Matthew Foster, Head of Leakage Strategy at Portsmouth Water, with HULO.
A deliberate trade off made years ago, for resilience and water quality, and true of a great many networks. An open network is not a backlog waiting to be finished.
A district metered area is measured at its boundary: water went in, less came out, and everything inside stays a single number. Half of this network has no boundary to measure at all.
HULO reads the pressure and flow monitors already installed, across zone boundaries and between them, and solves for the one place that explains every deviation at once. No hardware was installed for this engagement and nobody from HULO was on site. That is the starting point rather than the end of it: most utilities add sensors afterwards, once they can see which places are worth instrumenting.
These are the utility's own figures, not ours. Strategic Metered Areas of 30 to 60 km of main each, left open on purpose.
Every year. Delivering a 4.5 MLD reduction means finding thirty, not four and a half.
Worst case inside a large Strategic Metered Area, about a week on average. That is the cost of having no boundary to measure at.
For every megalitre a day missed against the target. Campaign work alone never closes that gap.
Two ways to close that gap: years of capital finishing the sectorisation, or more out of the data already being collected. The second had never been tested properly, so Portsmouth Water tested it rather than keep debating it.
The utility's leakage team wrote the conditions, on the grounds that most leak detection they had assessed assumed smaller DMAs would be built first.
Only the monitors already in the ground. The HULO team worked from the Netherlands throughout.
The utility's technicians opened hydrants at locations HULO was never given, at a realistic rate rather than a full open blast.
Six months of events against every repair job the utility's crews raised. Same zone, within 2 km of the three most likely cells, opened inside 30 days, strictly one to one. A looser rule would have given a bigger number and a weaker argument.
The full funnel, including the part that falls away.
Raised between March and August 2026, on the monitors already in the ground.
Carrying a location we are willing to defend. The rest had nowhere to send anyone.
Survived one to one matching against a repair work order the utility raised.
Generated by the leak itself. The other four are works signatures: the platform saw the repair, not the leak.
From the start of the HULO event to the moment the repair work order was opened. All fourteen ran ahead of the crew.
From the repair point, and 25 days before the job opened. Measured from the centre of the three most likely cells.
Which decides whether a crew is given a street or a zone.
Matthew Foster, Head of Leakage Strategy, Portsmouth Water
The least flattering measurement in the engagement, and the most useful. Nobody had measured it before, because nothing made it visible.
Of repair locations: the fleet as designed, against the fleet as it actually was on the day.
Landed on days when fewer than three sensors in the zone were sending trustworthy data.
Of paired repair points, even with every sensor healthy. Small leaks remain beyond this.
Two of the three cost almost nothing to fix. Knowing which one a network has is the question to answer before buying anything from anyone, including us.
Six pages: the method, the events that did not count, and the limits in full. Tell us who you are and it is yours.
Thank you. The case study is below, and a copy of the link is on its way to your inbox.
These films are a demonstration network, not Portsmouth Water's. We do not publish a client's live dashboard, their zone names or their work order numbers. The workflow, the evidence and the screens are the ones the utility used, and the story each film tells is the story that played out on their network. Silent, captioned, and nothing downloads until you press play.
Fleet health first: which sensors are flat, and which are sending a fraction of their readings. Then the blind spot map, shading the network by the largest leak that would go unnoticed in each cell, counting only the sensors actually reporting. This is the view that produced the 85 to 55 figure.
One case through the queue: the evidence that raised it, the harmless explanations ruled out, then every sweep the crew logged, street by street, until the leak was found. Not a coordinate. A location, a loss rate, and the argument behind both.
A field mechanic calls about one valve. The twin holds the network as it is running now, so the question goes to the model rather than to memory: close it, solve it, read back which nodes lose head and by how much. Mains stopped flowing, nothing below critical pressure, worst drop 24.1 m.
Five areas carry a steady excess above the demand they should show, and it holds through the overnight trough. Combined, about 61 L/s while drawing.
Roughly one megalitre a day as an upper bound, and only at the pattern's busiest. It is an order of magnitude, not a metered total, and it is not field verified. A missing logger, an unmetered supply and an illegal offtake all look like this, and they have very different answers. The full working is in the case study.
“You should not underestimate it. You can end up with really good software that nobody follows up on.”
Matthew Foster, Head of Leakage Strategy, Portsmouth Water
That is the part software does not solve. Alerts are worth something only if somebody owns them, has time to act, and feeds the result back in.
Find out what the network cannot see, then fix the cheap part first. That mechanism transfers. It is not a guarantee of a result, and a network cannot be read in ten minutes.
One week of pressure and flow, and a network model in whatever the GIS exports. About two hours of an engineer's time. Ten working days later: a loss map, the events HULO would have raised, and an honest read on what the instrumentation can support. If the answer is that you do not need us, we will say so in writing.
What we found, on which network, and what it cost to find it. Roughly monthly, and never a figure we cannot show you the working for.
HULO’s project Lekker (tegen lekken) is co-financed by the European Union, by SNN and by the Dutch Ministry of Economic Affairs.
