August 26, 2026 | Water • Wastewater • NRW • Metering • GIS • SCADA • Asset Management

This week’s idea: The next stage of utility AI is not another dashboard. It is connecting information already sitting in AMI, SCADA, GIS, work-order, billing, laboratory, and asset-management systems so operators can answer a much simpler question:

Where should we investigate today?

The newest AI models are becoming faster and less expensive to deploy. OpenAI’s GPT-5.6, released July 9, now includes multiple capability and cost tiers, while an Ultrafast API preview announced August 13 can run GPT-5.6 Sol at up to 14× standard processing speed.

For utilities, the important development is not simply a better chatbot.

Faster models make near-real-time operational assistants increasingly practical: analyzing meter exceptions, summarizing alarms, reviewing work orders, interpreting inspection reports, preparing operator shift summaries, or ranking field investigations.

The opportunity is to place AI above existing operational systems, rather than trying to replace them.

A July 2026 study in the Journal of Water Process Engineering used up to nine years of full-scale wastewater data and benchmarked 19 machine-learning models across 13 operational targets. Some models achieved R² values of 0.96–0.99 for variables including flow, energy, pumping, and pressure. The research also demonstrated AI-based “soft sensors” that can provide estimates when physical sensors fail or are unavailable.

That points toward an important shift:

SCADA tells you what is happening.
Predictive AI can help estimate what may happen next.

Potential applications include influent forecasting, pump-condition prediction, aeration optimization, chemical dosing, effluent-quality forecasting and early detection of abnormal process conditions.

Leak detection is moving from isolated pilots toward continuous network monitoring.

A recent Northumbrian Water deployment reports AI monitoring across 149 km of distribution network, identifying 130 investigation points and contributing to a reported leakage reduction of 267 m³/hour during the program.

Separately, Thames Water announced a 13-month AI-enabled satellite leak-detection program targeting more than 100 million litres per day of leakage reduction.

The lesson for NRW teams is important: AI does not replace leak-detection crews. It improves where those crews spend their time.

Yorkshire Water is expanding AI-enabled condition monitoring to another 1,200 wastewater pumping-station assets and 300 clean-water pumping assets. Its wider program already covers roughly 2,500 assets.

Dŵr Cymru Welsh Water also announced a three-year extension this month for predictive monitoring of critical wastewater equipment using electrical-signature analysis of pumps and motors.

This represents one of AI's clearest utility business cases:

Detect deterioration → prioritize maintenance → intervene before failure.

AI is simultaneously making sophisticated cyber capabilities easier to scale. Recent reporting highlights growing concern about AI-assisted attacks against critical infrastructure, including water systems.

That makes cybersecurity part of every utility AI strategy—not a separate IT discussion.

The United States Environmental Protection Agency is holding a Water Cybersecurity Assessment Tool webinar today, August 26, followed by cybersecurity procurement training on August 27.

Utility rule: never connect a generative AI application directly to SCADA control functions without appropriate architecture, segmentation, authorization, testing and human oversight.

Many utilities have plenty of data but still rely heavily on people manually deciding where to investigate.

Instead, create a daily AI-assisted ranking process.

Start with data you already have:

You do not need every dataset before beginning.

Start with four or five reliable sources.

For each account, route, DMA or pressure zone, calculate indicators such as:

Night-flow anomaly

Current minimum night flow ÷ normal minimum night flow

Consumption deviation

Current consumption − historical expected consumption

Meter-risk indicator

Age + estimated-read frequency + zero-read frequency + meter technology

Infrastructure risk

Pipe age + material + break history + pressure conditions.

Instead of asking:

ask AI:

That distinction matters.

AI becomes a decision-support layer, not an autonomous operator.

Example:

Now supervisors can start each morning with:

“Here are the five places most worth investigating today.”

This is where many AI projects fail.

Record what crews discover:

Leak confirmed?
Bad meter?
Vacant property?
Data error?
Normal operation?
Valve issue?
Service-line problem?

Those outcomes become feedback for improving the ranking model.

AI → Investigation → Field verification → Better AI

That feedback loop is more valuable than the algorithm alone.

Copy this prompt into your approved AI environment and provide a sanitized operational dataset.

The final sentence is surprisingly important. It forces the AI away from generic analysis and toward an operational decision.

ChatGPT can increasingly serve as an analysis layer across documents, spreadsheets, research and operational workflows rather than simply a writing assistant. The latest GPT-5.6 generation emphasizes professional knowledge work, tool use, analysis and multi-step workflows.

For utility professionals, practical applications include:

Metering: analyze AMI exceptions, estimated reads and consumption patterns.

NRW: combine production, consumption and field observations to prioritize investigations.

Asset management: summarize inspection records and identify recurring failure patterns.

GIS: help design geoprocessing workflows, SQL queries, Arcade expressions and Python scripts.

Engineering: review specifications, inspection reports and technical documents.

Customer service: classify customer complaints and summarize complex account histories.

Wastewater: analyze pump, flow, maintenance and process datasets.

Management: convert operational information into concise weekly reports and decision briefs.

The most important implementation principle remains:

AI should recommend. Authorized utility professionals should decide.

Greenville Water provides an excellent example of how utilities should approach AI.

A February 2026 Journal AWWA case study describes how the utility combined machine learning, AI approaches, mapping and institutional knowledge to improve leak-detection targeting, scheduling efficiency and main-replacement planning.

The technology is only part of the story.

Experienced operators already know things algorithms do not:

“Leaks frequently appear on this street.”

“This pressure zone behaves differently during summer.”

“These meters have historically caused problems.”

“This main looks acceptable in the database, but crews know otherwise.”

The strongest utility AI systems combine:

**Sensor data

Don't build AI outside operations.

Build AI with the people who already understand the system.

One of the most useful AI skills for utility professionals is learning how to force AI to distinguish between what the data actually shows and what the model thinks it might mean.

Consider:

Observation: Minimum night flow increased from 1.8 MGD to 2.3 MGD.

Poor AI conclusion: There is a major leak.

Better AI reasoning: Minimum night flow increased approximately 28%. Possible causes include distribution leakage, unusual legitimate demand, meter/data problems, operational changes or boundary-valve conditions.

Action: Compare production meters, pressure, AMI consumption, valve operations and recent work orders before dispatching leak-detection crews.

Use this structure in prompts:

OBSERVATION → POSSIBLE CAUSES → EVIDENCE NEEDED → RECOMMENDED ACTION

It dramatically improves the usefulness of operational AI outputs.

You do not need a digital twin.

You do not need an enterprise AI platform.

You need one spreadsheet.

Export one operational dataset.

Examples:

AMI reads
meter-reading exceptions
service orders
pump alarms
SCADA events
customer complaints
main breaks.

Remove unnecessary personally identifiable or security-sensitive information.

Keep only fields needed for analysis.

Ask your approved AI tool:

“Profile this dataset. Explain every field, identify missing data, duplicates, unusual values and potential operational KPIs.”

Ask:

“Identify the ten records or locations that deserve operational investigation and explain why.”

Ask:

“What additional dataset would most improve this analysis?”

This question is powerful because AI begins identifying your data integration roadmap.

It might recommend:

GIS.

Pressure.

Meter age.

Work orders.

Production.

Weather.

Previous failures.

Create one simple output:

Give it to an operator, supervisor or field technician.

Then ask:

“Does this list make operational sense?”

Their answer is your first AI model evaluation.

EPA's August cybersecurity program includes training on SCADA/ICS threats, its Water Cybersecurity Assessment Tool, cybersecurity procurement and a national exercise designed to test water and wastewater operations under degraded telecommunications and internet conditions.

The implication for AI programs is clear: digital transformation and cyber resilience must be designed together.

Rapid data-center expansion is creating new discussions around both electricity and water demand. Recent reporting estimates roughly 4,000 U.S. data centers with thousands more planned or under construction, increasing scrutiny of their resource requirements.

For water utilities, large AI/data-center developments may eventually affect demand forecasting, capacity planning, rate discussions, infrastructure investment and water-reuse strategies.

On August 19, EPA announced $10.8 million in available grant funding aimed at strengthening the drinking-water and wastewater workforce, citing aging infrastructure, retirements, evolving technology and cybersecurity among the challenges facing the sector.

AI adoption should therefore be viewed partly as a workforce-capacity strategy: capturing institutional knowledge, accelerating analysis and allowing experienced staff to concentrate on decisions that require utility expertise.

Don't start your AI program by asking:

“Where can we use artificial intelligence?”

Start with:

“What operational decision takes too long because the information is scattered across different systems?”

Then connect the information.

Metering may have AMI.

Engineering has GIS.

Operations has SCADA.

Maintenance has work orders.

Customer service has billing and complaints.

Laboratories have water-quality data.

Finance has costs.

The real opportunity is not creating another database.

It is creating intelligence across the databases you already have.

Start small.

Choose one decision.

Connect a few trusted datasets.

Use AI to rank—not automatically control.

Send the recommendation to an experienced utility professional.

Capture what happens in the field.

Feed that knowledge back into the process.

Repeat.

That is how an AI experiment becomes an operational system.

Artificial intelligence won't replace utility professionals. But utility professionals who learn how to combine AI with operational experience will increasingly redefine how utilities are managed.

The future utility will not simply collect more data.

It will get better at turning data into the next best action.