It’s easy to take water for granted. Turn on the tap, and you’ll receive clean, life-giving water (with some very notable exceptions). But for a myriad of reasons, ranging from our changing climate to aging infrastructure to growing demands for water, all aspects of the water cycle — how it is collected, cleaned, distributed (and repeat) — are overdue for a technological makeover.

For one thing, the workforce behind our waterworks is aging, at least within the public water utility sector, which is composed of an astounding 50,000 individual systems. “Lots of senior engineers are 30 years into their job and are reaching retirement,” says Will Maize, a water industry analyst with market research firm Bluefield Research. When they go, so will a good deal of institutional knowledge.

But as recent and prolonged droughts in the West reminded farmers, municipalities, and manufacturers, water scarcity calls for better water measurement and management. That’s why there’s an emerging shift toward what Will Sarni, CEO of the consultancy firm Water Foundry, calls “digital water.”

“When supply vastly exceeds demand, you can do stupid things, and we have a few hundred years of doing that,” says Sarni, referring to how utilities and the private sector have traditionally managed water.

Digital water is water that is managed using software-based tools such as data analytics, visualization, and predictive analytics. It goes by other monikers as well — Maize calls it “smart water.” Hardware, such as sensors to track water quality, pressure, and flow, is at the base of this new tech pyramid.

But to advance the digital analytics tools, a healthy dose of data science is needed. Sarni considers artificial intelligence to be “the holy grail” when it comes to digital water. With the right development, AI could unlock an incredible amount of value in terms of cutting waste, improving wastewater treatment systems, and keeping water infrastructure healthy.

Though most in the water industry agree that it’s still early days for AI tools that manage water. (Much earlier than for the energy sector.)

However, once water utilities, distributors, and companies catch on, AI and machine learning could transform water consumption and delivery systems. It’s an important problem to solve, as the world’s population grows to 9 billion by 2050 and water scarcity becomes of increasing concern.

Leaky Pipes and Mains

Rome is known for its aqueducts, which played an important role in expanding the Roman Empire. It’s ironic, then, that the city had to ration water during this drought-stricken summer, partly due to Rome’s notoriously leaky municipal water infrastructure that loses up to 44 percent of the water moving through the system.

According to industry group American Water Works Association, between 2011 and 2050, U.S. utilities will spend $1.7 trillion on repair and expansion of drinking water infrastructure, and more than half of that bill will go toward replacing distribution pipes as they reach their end of life.

But when does that time come? Replace them too soon, and you’ve squandered energy and resources. Wait too long, and pipes fail, wasting a great deal of clean water.

Fracta, a startup based in Redwood City, California, has developed a system that aims to help water utilities save money and resources by prioritizing cities’ replacement of water mains and distribution pipes, from which smaller pipes branch out into buildings.

Lars Stenstedt, COO of Fracta, says utilities are already looking to analyze as many of the variables that go into a given pipe’s life cycle as possible, but “as the number of relative variables go up, the only way to approach this is through machine learning.”

Fracta analyzes data related to the type of soil in which pipes rest, the topography, and weather records, and then applies machine learning to find patterns, across an entire city or region, that offer clues as to which pipes are at the greatest risk of leaks or failure. The company is currently working with two water utilities in the Bay Area to prove that its approach works.

Predictive analytics firm Pluto is applying artificial intelligence to the legacy water equipment and data management systems used by utilities and industrial water users, such as beverage manufacturers. The year-old company says it can help water managers conserve energy, lower operating costs, and squeeze more life from water plant assets. It does this by running data from disparate sources, such as pumps and filtration systems, through its algorithms to come up with optimal management and control protocols.

Wastewater treatment systems filter water through polymer membranes with very tiny pores, but these filters need to be regularly backwashed or chemically cleaned to keep the water flowing and extend the filter’s life cycle. Water Planet, a company that sells these filtration systems, has developed machine-learning software that continually analyzes data from flow and pressure sensors as well as large historical data sets to determine the optimal filter backwash protocol, based on both current water quality and past events and patterns.

Water Planet sells the tool, called IntelliFlux, to industrial water plant managers but wants to expand to utilities, which have larger systems and seasonal variability, such as an uptick in polluted water during heavy rainstorms. Jason Lake, marketing director at Water Planet, says incumbent filter-cleaning systems also monitor flow and pressure, but without artificial intelligence, “they can’t say what [incoming] water [quality] will be in the future, just know what it was in the past.”

Water Use in Homes and Buildings

In more than half of all U.S. households, electric and gas utilities have swapped out analog meters for “smart” meters that transmit data wirelessly, making it easier to collect usage data and enable a raft of energy-saving programs. But so far, only a small percentage of water meters are smart.

That’s soon to change, because utilities want to take advantage of AI-based data tools that work atop smart meter networks.

Arizona-based software company Fathom manages water billing systems and analyzes data related to its customers—from their employment status to sentiments they express on social media—to identify those who are unlikely to pay their bills on time. Fathom is also starting to apply AI toward crunching smart meter data to identify water leaks.

“Say you have 100,000 meters, and each one is using water in slightly different ways,” explains Graham Symmonds, chief knowledge officer at Fathom. Differentiating a household that has developed a leak from one that has consistent water demand — because, for example, its swimming pool pump is always running — requires advanced pattern recognition. Without that, utilities just look for meters that never show zero flow and assume it signifies a leak, but that generates a lot of false positives.

San Francisco–based Valor Water Analytics has developed machine-learning algorithms to detect inaccuracies or anomalies in meter data. It does so to help locate where consumed water may have gone unbilled, which costs utilities billions in lost revenue. But Valor also provides utilities with tools they can use to work with customers to reduce water use and identify leaks.

“We are working to raise the bar for more efficient and proactive meter, revenue, and customer account management,” says Christine Boyle, founder and CEO of Valor. “AI allows us to understand universal versus local trends.”

Of course, it’s not just startups looking to apply AI to the water sector. General Electric, Siemens, and other stalwarts of industrial control systems have developed AI-based software specifically for water managers.

During California’s recent draught, OmniEarth, a geospatial data analytics provider, put IBM’s Watson AI platform to work, analyzing consumption data provided by utilities and satellite and aerial imagery. OmniEarth then worked with utilities to communicate with specific customers — like those who have swimming pools, for example — to suggest steps they could take to reduce consumption in line with the state’s conservation goals.

Within the water sector, the emergence of AI-based data tools is part of a shift toward more proactive operations, based on “algorithms that give you insights from data,” explains Maize. Water managers really need any advantage they can afford, he adds, because traditionally water utilities are incredibly reactive.

We are also facing more frequent, less predictable, and more intense storms, as well as extended periods of drought, so water utilities can’t afford to stay reactive and still meet their mandates to provide clean water to a growing population.

Sarni cautions that the sector won’t modernize overnight: “We’re in our infancy in terms of thinking about how AI can play in the world of water.” AI is a generating a great deal of buzz, but “we need to think of this as a journey.”