If you’ve been to a water conference recently, you’ve probably noticed that "Artificial Intelligence" has officially dethroned "Digital Twin" as the industry’s favorite buzzword. Vendors are slapping "AI-powered" onto everything from SCADA dashboards to turbidity sensors.
But if we are being completely honest with ourselves, the water sector’s current relationship with AI is a lot like my relationship with my gym membership: highly aspirational, poorly understood, and barely yielding results.
We are making some hilarious and expensive mistakes. Let’s look at where we are going wrong, and more importantly, how to stop making a splash in the shallow end of the tech pool.
Mistake #1: The "psychic plumber" syndrome
Right now, utility executives are buying AI platforms hoping they will act as a magic wand for 50-year-old infrastructure. There is a bizarre assumption that if you plug an algorithm into your network, it will miraculously fix your non-revenue water (NRW) problem.
Newsflash: AI cannot tighten a loose valve. If your distribution network is held together by rust, duct tape, and the sheer willpower of your lead operator, AI is just going to give you a highly accurate, color-coded countdown to your next catastrophic main break.
What to do instead: Treat AI as a navigational tool, not a repairman. Use it to prioritize your Capital Improvement Plan (CIP). Instead of saying, "AI, fix my leaks," ask, "AI, based on our pressure data and acoustic sensors, which three miles of pipe should we replace this year to avoid ending up on the evening news?"
Mistake #2: The data swamp (garbage in, garbage out)
Machine learning models are hungry. They need data to learn. But what happens when the diet you feed your million-dollar AI consists of SCADA data that drops offline every time it rains, handwritten operator logs from 1998, and Excel spreadsheets where half the cells just say #DIV/0!?
You get Artificial Stupidity.
Many utilities try to implement predictive maintenance models without first cleaning up their data architecture. You cannot predict when a pump will fail if your sensor calibration has been drifting since the Obama administration.
.jpg)
What to do instead: Get your data house in order before inviting the AI guests over. Invest in data standardization and clean, reliable telemetry. Start with a "data readiness" audit. If your human engineers can't make sense of your historical data, the algorithm won't be able to either.
Mistake #3: Man vs. machine paranoia
There is a lingering fear in the control room that AI is Skynet, and it’s coming for Bob the Operator’s job. Because of this, when AI recommends dropping the coagulant dose by 10% to save money, Bob ignores it and does what he’s done every Tuesday for the last twenty years.
Let’s clear the air: AI is not coming for Bob’s job. AI doesn't know how to jiggle the handle on the clarifier to make it stop humming. AI just wants to automate the soul-crushing math of energy optimization.
.jpg)
What to do instead: Focus on Augmented Intelligence, not Artificial Intelligence. Bring your operators into the AI procurement and training process from Day 1. Frame technology as an assistant that takes care of the boring, complex calculations so the operators can focus on the actual plant strategy. When the algorithm and the operator trust each other, that’s when you actually see ROI.
The bottom line
AI is an incredibly powerful current in the water sector, and we are right to be excited about it. It can optimize pump scheduling to save millions in energy costs, predict membrane fouling before it happens, and radically reduce chemical usage.
But it is not magic. It requires clean data, realistic expectations, and human collaboration. If we stop treating it like a magic plunger and start treating it like the high-powered analytical tool it is, we might actually get our heads above water.
