Optimization

Find the water you already lose before you pay for new supply.

We think a utility should find the water it already loses before it pays for new supply. Lost water is paid for twice, once to pump and treat it and again to replace it. WHO and UNICEF reported in August 2025 that 2.1 billion people still lack safely managed drinking water. A World Bank study, cited by the Bank in 2016, put the water lost each year to physical leaks from supply networks at 32 billion cubic meters, half of it in developing countries. extendfuture would model the whole water system, so a utility can see which pipes, pumps and reuse projects return the most water for the money.

Explore with AI

ChatGPT Perplexity Claude Google Grok

How can a water utility reduce leakage with AI?

Start with a hydraulic model of the pipe network, then use AI to work through it at scale. We would build the model in EPANET from pipe records and meter data, and adjust it until it matches measured pressures and flows. Comparing flows at night, when use is lowest, shows districts where water goes missing. Optimization then ranks pipe replacement and pressure changes by water saved per unit of cost. AI agents, software that sets up runs and drafts summaries, compare the scenarios. Engineers confirm leaks on the ground and sign off the plan.

Fix losses before you add supply

We would start every water plan with the water already lost. The World Bank calls water that is pumped and then lost or unaccounted for non-revenue water, and says that where water is scarce, managing it is often more cost-effective than adding supply.

Wastewater is the other half of the system. Of the 332 billion cubic meters of household wastewater generated in 2024, 185 billion were safely treated, according to the UN's 2026 SDG report. The rest is discharged without safe treatment, and it is also water that could be treated and reused.

Water needs the utility's records from the start

Open data is not enough here, because the pipe network sits in the utility's own records. The model would cover sources, treatment plants, pumps, tanks, pipes and demand by district, and the wastewater that comes back.

Pipe records with material, diameter and age from the asset register
Pressure and flow readings from district meters and pumping stations
Billing totals by district, to compare water supplied with water billed
Rainfall, terrain and population data from open and official sources, to forecast demand
Repair and burst history, to estimate which pipes are likely to fail next

Treat a mismatch with the meters as a lead

Once the EPANET model is calibrated to the meters, a district where the model and the meters keep disagreeing is a lead. Either the records are wrong or water is leaving the pipes. Field crews confirm or reject each lead, and their findings go back into the model. Demand forecasts are checked against what happened in past years, including a dry one, before they are used to size anything.

Spend repair money where it saves the most water

Repairs should be ranked by water saved per unit of cost, in a plan that utility managers and regulators can defend.

Districts ranked by estimated loss, with the evidence behind each estimate
A pipe replacement list ordered by water saved and failure risk per unit of cost
Pressure settings that would reduce leakage while keeping minimum pressure at every tap
Pump schedules that shift energy use to cheaper hours where tanks allow
Options for treating and reusing wastewater, with the demand each could replace

Operators stay in charge of pumps and valves

Recommendations about pressure or pumping go to the control room as proposals, and operators decide whether to apply them. Billing data is used as district totals, and no one sees a household's use without a reason to. Fairness checks show which districts get repairs first and why, so the order can be defended in public. Every run is logged with its data and version, and any engineer can rerun it.

A utility should own its network model

A utility should be able to keep its model and share it with its regulator. Each tool below is open source or public domain.

EPANET, the US EPA's public domain software for modeling water flow, pressure and quality in pipe networks
WNTR, an EPANET-compatible Python package for simulating water networks under disaster scenarios
SWMM, the EPA's open model for runoff and for stormwater, combined and sanitary sewers
QGIS, the open-source geographic information system, for maps of pipes, districts and leaks
statsforecast, a library of statistical forecasting models, for water demand
OR-Tools, for ranking repair work and scheduling field crews

We do not have a complete pipe map. Can we still start?

Yes. The first model uses the records that exist and flags the districts where missing data would change the answer, so field teams know which records to check first. The model improves as the records do.

Will this find individual leaks?

It narrows the search. The model points to districts and pipe sections where water goes missing. Crews then use listening equipment or other field methods to find the exact spot.

Can it plan for drought?

Yes, as scenarios. Demand and supply limits can be run for a dry year and a typical one. The report shows when tanks and sources would run short and which measures, such as pressure management or reuse, buy the most time.

Does the model control pumps or valves?

No. It proposes settings and schedules, and the existing control systems stay in charge.

How would wastewater reuse be planned?

SWMM models the sewers, and each treatment and reuse option enters the optimization with a cost and a quality limit. The report shows how much demand treated wastewater could replace, such as for industry or irrigation, and at what cost.

Sources

Talk to the people who build.

One call. An honest read on what AI can do for this, and the number it has to beat.

Start a project Book a call