Proof · HULO

Nama Water Services · Ghala, Oman

One working sensor
in five.

The hardest starting conditions of any deployment here, and the tightest result of any deployment here. Delivered with Witteveen+Bos and BlueGold Technology.

Localisation

250 × 250 m

The ground a crew has to cover. It holds six tenths of one per cent of the mains in the zone, so 99.4% of them are never walked.

Time to detection

15 min

From system activation to the first detected event.

Loss quantified

1,000 m³/day

Measured, not estimated from a water balance.

Starting point

1 of 5

Pressure sensors actually working in the region.

One functioning pressure sensor out of five installed. District metered areas that could not be isolated. Problems inside the reservoirs, including float valves found disconnected. A network topology that resisted straightforward monitoring.

Nothing was added to fix it. Sensor positions were chosen using control-systems engineering rather than convenience, eighteen locations, and integrated with the existing SCADA.

Why this is the case we put first

Any vendor can show a good result on a well-instrumented network. Oman is the opposite: the worst starting conditions on this page produced the tightest localisation on this page.

The argument running through this whole site. That the signal is usually already there and simply is not being read, is easy to dismiss until you watch it work on a network where four sensors in five were dead.

Aegea / Águas do Rio · Rio de Janeiro

From 1 in 10
to 9 in 10.

The largest test of the argument on this website, in a network with no district metered areas at all, and no plan to build any. Presented on stage by HULO and Aegea together, which is the only reason these numbers are publishable.

Detection rate

1 → 9 in 10

One in ten leaks found at the start of the programme. Nine in ten by the end.

Search area

500 → 2 km

Often 200 to 600 metres, in Aegea's own account of the field work.

Detection time

< 2 hours

From event to a prioritised area a crew can be sent to.

Sensors used

300 → 100

Fewer sensors, not more. Repositioned and repaired rather than added to.

Hundreds of sensors, data available, real ambition. And underneath: some sensors not working, some installed somewhere other than where the records said, some reporting once a day. Search areas running to 500 kilometres.

“More data did not solve the problem. Better understanding did.” Robbert Lodewijks, CEO and co-founder, said that on stage in Rio. And from Aegea, in their own words: more sensors do not guarantee better results; understanding your network does.

Read this one carefully

These results come from controlled tests, not real leaks. Moving to live events and scaling across the network is the next phase. We would rather say that here than have you discover it in a reference call.

Águas do Rio operates without DMAs and does not intend to build them. That is exactly why this test matters: it asks whether a utility can gain visibility without restructuring the network first. On controlled events, the answer was yes.

Portsmouth Water · Binsted Ford & Yapton · May 2026

Ten flushes.
Nine inside the target.

Portsmouth opened ten hydrants at locations they did not disclose, across two areas, under deliberately varied conditions, one with a logger missing, two overlapping, one timed to collide with the morning demand rise.

Inside the agreed target

9 of 10

Across both areas, flushes that landed inside the search area Portsmouth agreed beforehand: no larger than a quarter of the main area. Four of the five at Binsted Ford. These are the results from the data; read the limitation below for what that means in real time.

Tightest result

1.4%

Of the main area, at Binsted Ford: a 98.6% reduction in ground to cover. The others in that area came in at 7.5%, 13.6% and a very small zone.

Outside target

27%

The one flush of the ten that missed, with a second candidate area about 500 m away.

Programme

3 of 5

Controlled flush testing is phase three of five, not the end goal.

At Yapton, one flush localised correctly even though a pressure logger had been removed by mistake before the test. Two flushes thirty minutes apart in a shared window were both captured, and as it happened one of the two could not be told apart from the other. In the data afterwards it could. That is the honest shape of these tests: they were deliberately messy, and separating them live is harder than separating them later.

That one is worth understanding rather than explaining away: a ninety-minute flush timed into the morning consumption rise looks like the morning consumption rise. A real leak does not stop after ninety minutes. It persists, and stays visible once the rise settles.

The limitation this test exposed

Four of the Binsted Ford flushes ran back to back overnight. In real time HULO captured them as one long combined event, not four. The individual accuracies above come from separating that event afterwards and re-running the analysis per flush.

So read them correctly: they show the localisation model is accurate, and that real-time event separation was the weakness when interventions land minutes apart in the same neighbourhood. Only the fifth flush, which stood alone, was both detected and localised live.

Podcast · The Executive Exchange · Isle

Navigating Networks: DMAs and the New Age of Water Management

Bob Taylor, CEO, Portsmouth Water

Interviewed by Piers Clark · 13:19

00:00 / 13:19

WML & Gruppo CAP

Named, without
a number.

Two engagements where the work is real and the measurement is not finished. They appear here anyway, because leaving them off would be its own kind of dishonesty, and they appear without figures, because none have been agreed.

WML · Limburg, The Netherlands

A public utility, speaking carefully.

Public utilities do not endorse vendors. What WML will say is what the work is for: an additional analytical layer on the operational data they already collect, supporting the people who interpret it.

“Understanding how the network behaves in different situations helps us interpret the data we already collect. Software that provides additional insight can support operational awareness and help teams investigate events more effectively.”

Rolf Cuijpers, process operator, Waterleidingmaatschappij Limburg

Gruppo CAP · Milan, Italy

The instrumentation is already bought.

Italy spent its recovery-fund money well: sensors, district metering, digital infrastructure, data collection. Gruppo CAP now has an instrumented network and the question every utility reaches next, what does all of it actually tell us on a Tuesday morning?

At assessment stage: anomalies detected on the existing network and validated against what the field found.

Why Italy is different right now

Most markets have to be persuaded to invest in visibility first. Italian utilities already have. The gap that remains is not more measurement. It is understanding what has been measured. NIS2 has also been enforced there since October 2024.

Beyond leak detection

What is happening?
Why does it matter?
Where to act?

Leakage is where a utility starts with us, because it is urgent and it is measurable. It is not where HULO stops. Everything below is already running on the wider Portsmouth network, outside the controlled tests, on real behaviour. It is also what the Rio deployment with Aegea and Águas do Rio runs on day to day, and what the Milan engagement is scoped around.

What is happening?

LIVEReal network eventsAbnormal behaviour surfaced beyond controlled tests, the network telling you something without being asked.

LIVEPressure intelligenceLive pressure awareness across operational areas, not a reading at a point.

LIVEArea snapshotsReal-time operational state, per zone, at a glance.

Why does it matter?

LIVELarge consumer visibilityUnusual demand behaviour and what it costs operationally, often not a leak at all.

LIVEForecasting behaviourExpected against abnormal conditions, so a deviation has something to be measured against.

Where to act?

LIVESensor health monitoringBlind spots identified before they become operational risk. Knowing what you cannot see is itself an answer.

LIVELeak localisationThe wedge: an area a crew can walk, ranked by what it is costing you.

NOT A ROADMAP, NOT A PROMISE, AND NOT SOMETHING WE NEEDED A CONTROLLED TEST TO DEMONSTRATE.

This is the difference between a leak detection tool and an intelligence layer.

A tool answers one question. A layer answers the question you had, and then the two you had not thought to ask, because it is reading the same network, continuously, with a model of how it ought to behave.

It is also why the sensor health work matters as much as the detection work.

Software that cannot tell you where it is blind is not giving you intelligence.
It is giving you output.

The first question everybody asks

How many sensors
does it take?

Fewer than you would expect, and more of them buys you a smaller search area. Six zones we run, from one sensor every 41 kilometres to one every 3.6, with the area a crew was given at the end of it. Every one of them detected, classified and localised in under fifteen minutes.

Where	Setup	Sensors	Network	One sensor per	Search area	Of the network
The Netherlands	Open network	1 flow · 10 pressure	450 km	41 km	900 × 900 m	1%
Latin America	Open network	2 flow · 17 pressure	321 km	17 km	500 × 500 m	0.8%
Italy	Pressure zone	3 flow · 16 pressure	203 km	11 km	500 × 400 m	0.6%
The Netherlands	Pressure zone	4 flow · 18 pressure	100 km	4.5 km	200 × 200 m	0.2%
Italy	DMA	3 flow · 3 pressure	25 km	4.2 km	400 × 200 m	0.6%
Middle East	DMA	2 flow · 16 pressure	65 km	3.6 km	250 × 250 m	0.6%

There is no minimum. There is a trade.

The sparsest zone in this table carries one sensor for every 41 kilometres of main, and it still put a crew inside a 900 metre square. Nothing here needed a network to be instrumented first. What the closer spacing buys is a smaller box: at one sensor every four kilometres it is 200 metres, which is a street rather than a neighbourhood.

Which of those is right for you is a question about what a search costs you, not about what the software needs. The feasibility scan answers it on your own network, and where the instrumentation cannot carry what you want, it says so before you sign anything.

Starting from nothing

No sensors yet?
Start with where.

A hydraulic model is enough to begin. Give us the model and no measurements at all, and the first thing HULO answers is not where the leaks are — it is where the sensors should go to find them. That is a smaller, cheaper question than instrumenting a network and hoping.

First, connect

Whatever you already measure, however little, plus the model. This is the step that costs nothing and settles what is possible.

Then the network tells you the next move

Sometimes it is a sensor, in a place chosen because of what it would reveal. Sometimes it is a repair that is already findable and has not been done. The point is that there is a next step, and it is a specific one.

And it repeats

Every improvement changes what the next best move is. Non-revenue water comes down as a sequence of decisions rather than as one purchase, and the software's job is to keep naming the next one.

What we learned

Two capabilities. Two limits.

A utility deciding on continuous monitoring needs the limits more than the wins. Both are here, in the same size type.

01Localisation is strong. Given a clean event window, HULO converges to as tight as 0.6% of the original network, or 1.4% of a main area.

02It survives poor conditions. Four dead sensors in five, a logger removed mid-test, overlapping events. None of it stopped the model converging.

03Closely spaced events merge. Back-to-back interventions in one neighbourhood were aggregated into a single event in real time. Narrower windows separate them.

04Short transients can hide. A ninety-minute event during the morning rise is hard to distinguish. Persistent losses, the ones that cost water, are not.

Start a feasibility scan Open the Proofroom

Five networks.
Every result.

One working sensor
in five.

From 1 in 10
to 9 in 10.

Ten flushes.
Nine inside the target.