Across North America, water utilities face a difficult financial and operational reality: unplanned repairs on aging infrastructure run anywhere from two to five times the cost of scheduled maintenance. And those emergency bills keep climbing as labor and materials grow more expensive.
There is also a critical infrastructure crisis. Because equipment installed during the post-war boom is past its intended life, breakdowns are happening more often – roughly 260,000 water main failures a year at a direct repair cost of about $2.6 billion in the United States and Canada alone, according to the American Society of Civil Engineers.
For finance and management, every unplanned shutdown carries immediate operational expenses and can complicate long-term rate planning. Many utilities have turned to AI-based Predictive Maintenance, yet results have often fallen short of expectations.
Compounding the issue is a looming “silver tsunami”: nearly one-third of experienced operators and maintenance staff will reach retirement age within the next decade, shrinking the pool of people who know these systems inside out.
This article explores a pragmatic alternative that acknowledges shortages in skilled labor.
It proposes a hybrid model that combines AI with the domain expertise now edging toward retirement. It’s about AI-generated failure signals that are filtered through human judgment. The opportunity is to create asset failure alerts (the engine of Predictive Maintenance) with high confidence and the operational context needed for timely, targeted action.
The Case for AI-Based Predictive Maintenance
Almost every industry is exploring new ways to apply AI – whether to boost efficiency, reduce costs, or unlock insights that were previously out of reach.
Water utilities are no different; in fact, the potential value becomes even clearer when one considers their operating reality. Within plants there can be large numbers of pumps, and motors spread across wide service areas. They are often in places that are out of reach and are not easily visible.
AI-enabled Predictive Maintenance offers the scale that people alone cannot: sensors stream vibration and pressure, and Machine-Learning models analyze data for the signatures that precede a breakdown.
Advance notice matters: shutting a pump down at 3 p.m. for a planned seal replacement avoids the overtime, water loss, and customer complaints that come with a 3 a.m. emergency call-out.
Done well, the payoff is material. The U.S. Department of Energy reports that a mature predictive-maintenance program can reduce 8-12 percent of total maintenance spend compared with standard preventive schedules. It’s not only about maintenance costs but also extending asset life.
From a strictly monetary viewpoint, Predictive Maintenance becomes compelling when three streams of value outweigh capital and change-management costs:
Reduced emergency spending – the largest, fastest-realized benefit and frees cash immediately.
Extended asset life defers replacement projects, smoothing capital-expenditure curves and lowering debt requirements.
Productivity gains allow a smaller workforce to manage a growing asset base, flattening future labor costs.
Utilities that pilot carefully and document those gains often report simple payback periods of two to four years. Failure to involve finance early, however, risks under-budgeting for sensors, IT integration, or staff training, pushing ROI out beyond planning horizons.
Limits of Pure-Play AI Predictive Maintenance
Despite the promise of AI based Predictive maintenance, there is one major drawback – False Positives.
At the core of pure-play AI based Predictive Maintenance platforms are learning algorithms that have been trained to analyze vast amounts of data generated from the sensors embedded in plant equipment and identify abnormal data patterns.
This is where False Positives come in: algorithms misclassify normal operating behavior as suspicious, thereby alerting operators to a potential asset failure. Reliability and maintenance crews are diverted to investigate and remediate when equipment is running normally.
This is because algorithms are often agnostic or blind to critical context such as seasonal operating changes, historical workarounds, or recent manual adjustments which are not captured by that a system level.
A Hybrid Predictive-Maintenance Model
The optimal predictive-maintenance strategy modified the Pure-Play AI approach with domain expertise at two points:
Up-front training. In the case of rotating equipment – pumps, aerators, blowers – the model is first exposed to the vibration and power-draw patterns that historically precede real faults in water and wastewater service. This targeted training grounds the algorithm in the failure modes that matter and eliminates the long, “learn-as-you-go” benchmarking phase that often produces False Positive alerts.
Ongoing validation. Each alert of potential asset failure must be reviewed by a knowledgeable analyst who either confirms or dismisses it and feeds that decision back into the model. With every cycle the algorithm becomes more accurate, steadily reducing false positives -the very issue that generic, black-box platforms struggle to control.
By combining AI based detection with continuous human feedback, utilities can receive early, trustworthy warnings and then schedule repairs instead of reacting to surprise breakdowns.
Going Forward
Will AI solve every systemic weakness in North American water utilities? No.
It cannot substitute decades of field experience or close the funding gap for pipe replacement. But ignoring predictive analytics leaves operators paying an ever-higher penalty for surprise failures. The prudent path is neither all-human nor all-machine; it is a deliberate blend of machine intelligence and human expertise. Executed thoughtfully, that hybrid model turns reactive losses into planned investments, preserves institutional knowledge, and steadies long-term finances.
Gilad Horn is the CEO of Israeli technology company Aquatis. He is a seasoned entrepreneur and strategic technology consultant with more than 20 years of experience in the industrial, governmental and startup sectors.
Explaining the Decline in WIFIA Loan Volume: Part 2
December 8, 2025
Commentary: Ductile Iron Provides Owners with Safety, Longevity
June 10, 2025
Commentary: 5 Tips for Navigating Federal Funding Chaos
April 7, 2025
By Gilad Horn
Across North America, water utilities face a difficult financial and operational reality: unplanned repairs on aging infrastructure run anywhere from two to five times the cost of scheduled maintenance. And those emergency bills keep climbing as labor and materials grow more expensive.
There is also a critical infrastructure crisis. Because equipment installed during the post-war boom is past its intended life, breakdowns are happening more often – roughly 260,000 water main failures a year at a direct repair cost of about $2.6 billion in the United States and Canada alone, according to the American Society of Civil Engineers.
For finance and management, every unplanned shutdown carries immediate operational expenses and can complicate long-term rate planning. Many utilities have turned to AI-based Predictive Maintenance, yet results have often fallen short of expectations.
Compounding the issue is a looming “silver tsunami”: nearly one-third of experienced operators and maintenance staff will reach retirement age within the next decade, shrinking the pool of people who know these systems inside out.
This article explores a pragmatic alternative that acknowledges shortages in skilled labor.
It proposes a hybrid model that combines AI with the domain expertise now edging toward retirement. It’s about AI-generated failure signals that are filtered through human judgment. The opportunity is to create asset failure alerts (the engine of Predictive Maintenance) with high confidence and the operational context needed for timely, targeted action.
The Case for AI-Based Predictive Maintenance
Almost every industry is exploring new ways to apply AI – whether to boost efficiency, reduce costs, or unlock insights that were previously out of reach.
Water utilities are no different; in fact, the potential value becomes even clearer when one considers their operating reality. Within plants there can be large numbers of pumps, and motors spread across wide service areas. They are often in places that are out of reach and are not easily visible.
AI-enabled Predictive Maintenance offers the scale that people alone cannot: sensors stream vibration and pressure, and Machine-Learning models analyze data for the signatures that precede a breakdown.
Advance notice matters: shutting a pump down at 3 p.m. for a planned seal replacement avoids the overtime, water loss, and customer complaints that come with a 3 a.m. emergency call-out.
Done well, the payoff is material. The U.S. Department of Energy reports that a mature predictive-maintenance program can reduce 8-12 percent of total maintenance spend compared with standard preventive schedules. It’s not only about maintenance costs but also extending asset life.
From a strictly monetary viewpoint, Predictive Maintenance becomes compelling when three streams of value outweigh capital and change-management costs:
Utilities that pilot carefully and document those gains often report simple payback periods of two to four years. Failure to involve finance early, however, risks under-budgeting for sensors, IT integration, or staff training, pushing ROI out beyond planning horizons.
Limits of Pure-Play AI Predictive Maintenance
Despite the promise of AI based Predictive maintenance, there is one major drawback – False Positives.
At the core of pure-play AI based Predictive Maintenance platforms are learning algorithms that have been trained to analyze vast amounts of data generated from the sensors embedded in plant equipment and identify abnormal data patterns.
This is where False Positives come in: algorithms misclassify normal operating behavior as suspicious, thereby alerting operators to a potential asset failure. Reliability and maintenance crews are diverted to investigate and remediate when equipment is running normally.
This is because algorithms are often agnostic or blind to critical context such as seasonal operating changes, historical workarounds, or recent manual adjustments which are not captured by that a system level.
A Hybrid Predictive-Maintenance Model
The optimal predictive-maintenance strategy modified the Pure-Play AI approach with domain expertise at two points:
Up-front training. In the case of rotating equipment – pumps, aerators, blowers – the model is first exposed to the vibration and power-draw patterns that historically precede real faults in water and wastewater service. This targeted training grounds the algorithm in the failure modes that matter and eliminates the long, “learn-as-you-go” benchmarking phase that often produces False Positive alerts.
Ongoing validation. Each alert of potential asset failure must be reviewed by a knowledgeable analyst who either confirms or dismisses it and feeds that decision back into the model. With every cycle the algorithm becomes more accurate, steadily reducing false positives -the very issue that generic, black-box platforms struggle to control.
By combining AI based detection with continuous human feedback, utilities can receive early, trustworthy warnings and then schedule repairs instead of reacting to surprise breakdowns.
Going Forward
Will AI solve every systemic weakness in North American water utilities? No.
It cannot substitute decades of field experience or close the funding gap for pipe replacement. But ignoring predictive analytics leaves operators paying an ever-higher penalty for surprise failures. The prudent path is neither all-human nor all-machine; it is a deliberate blend of machine intelligence and human expertise. Executed thoughtfully, that hybrid model turns reactive losses into planned investments, preserves institutional knowledge, and steadies long-term finances.
Gilad Horn is the CEO of Israeli technology company Aquatis. He is a seasoned entrepreneur and strategic technology consultant with more than 20 years of experience in the industrial, governmental and startup sectors.
Share this post