August 19, 2026 | Weekly Edition

Water utilities have invested heavily in meters, AMR, AMI, billing systems, GIS and customer information systems. Yet installing more technology does not automatically reduce Non-Revenue Water (NRW).

The real business opportunity is turning the information those systems already collect into better decisions.

Think about a utility with 600,000 customer accounts. A manager does not need another spreadsheet containing 600,000 rows. The useful question is:

Which 100 accounts should my team investigate first?

That is where AI becomes valuable.

AI can help utilities identify unusual consumption, recurring estimated reads, aging meters, communication failures and other patterns that may point to apparent losses or billing problems.

The objective isn't to replace meter readers, billing staff or NRW professionals.

It is to help them focus limited resources where they can have the greatest impact.

NRW is often discussed as a water-loss problem.

It is also a business-performance problem.

A utility can produce and deliver water successfully and still fail to bill accurately for all of it.

Revenue can be affected by inaccurate meters, estimated readings, incorrect account information, meter communication problems, billing errors and unrecorded consumption.

Individually, these problems may appear small.

Across tens of thousands of accounts, they can become significant.

"How many meters did we replace?"

to:

"How much water and revenue did our meter program recover or protect?"

That is a much stronger measure of performance.

One of the most important developments isn't a new AI model. It is the continued movement toward connecting GIS, operational data and AI.

Esri recently described its direction for agentic AI in GIS: AI agents that can work with authoritative maps, layers and spatial data rather than relying only on generic language-model knowledge. (Esri)

For water utilities, the business implication is significant.

Imagine asking:

"Show me the neighborhoods with the highest concentration of estimated reads."

Then:

"Which of those areas also have older meters?"

Then:

"Prioritize the locations where meter replacement could have the greatest impact."

GIS provides the where.

Meter and billing systems provide the what.

AI can help determine where to investigate first.

This is a practical AI workflow almost any utility can test.

Start with one billing cycle rather than the entire system.

Take an anonymized export containing basic information such as account ID, meter type, meter age, read type, current consumption, previous consumption and recent estimated-read history.

Ask an approved AI or analytics tool to separate the accounts into three groups:

Priority 1 : Investigate now

Repeated estimates combined with significant historical consumption or another strong business indicator.

Priority 2 : Review

Accounts showing emerging patterns that may require attention.

Priority 3 : Monitor

Isolated exceptions without enough evidence to justify immediate field work.

Now your team isn't starting with thousands of accounts.

It is starting with a prioritized worklist.

Better prioritization can mean fewer unnecessary truck rolls, better use of meter technicians, faster resolution of billing problems and greater opportunity to recover revenue.

AI hasn't made the final decision.

Your staff has simply started the day with better information.

It doesn't ask AI to diagnose the problem.

It asks AI to prioritize the investigation.

That distinction matters.

For utilities already using Esri's ArcGIS, GIS can become much more than an asset map.

Esri describes modern water GIS as supporting asset management, operational intelligence, engineering and customer care while incorporating real-time information and AI/ML capabilities. (Esri)

For an NRW manager, imagine a map showing:

Estimated reads + meter age + consumption anomalies + service orders + known leaks

Instead of reviewing each problem independently, managers can see where multiple issues are concentrated.

That can turn GIS into a management decision tool, not simply a mapping system.

A useful lesson this week comes from wastewater rather than drinking water.

The Central Valley Water Reclamation Facility has been working on AI readiness by first improving the foundation needed to support AI.

The approach is noteworthy because the facility did not begin by asking:

"Which AI platform should we buy?"

It focused first on data quality, infrastructure, workforce readiness and identifying operational problems worth solving. (Water Online)

This applies directly to NRW.

Before purchasing another AI platform, ask:

Is our meter data reliable?

Are meter replacements updated correctly?

Can billing, GIS and meter information be connected?

Do we know what business problem we want AI to solve?

Bad data does not become good data because AI analyzes it.

One of the easiest ways utility leaders can improve their use of AI is to change the questions they ask.

Don't ask:

"Analyze this meter report."

Ask:

"Identify the five issues management should address first, explain why they matter financially or operationally, and recommend the next action."

The second question forces AI to think in terms of business priorities.

That is much closer to how managers actually make decisions.

The same approach works across the utility:

Asset management: Which assets deserve inspection first?

Customer service: Which recurring complaints deserve management attention?

Wastewater: Which recurring maintenance issues create the greatest operational risk?

Engineering: Which projects should receive further evaluation?

NRW: Where should our field teams investigate first?

This week's challenge is intentionally simple.

Take one billing cycle.

Use anonymized data.

Ask an approved AI or analytics tool to identify the 10 accounts with the strongest combination of repeated estimated reads and significant historical consumption.

Then have someone who understands the accounts review the results.

Don't launch a major AI project.

Don't buy new software.

Don't build a complicated model.

Spend 30 minutes answering one question:

If the answer is yes, you have the beginning of a business case.

A newly published July 2026 study demonstrated how machine learning can support full-scale wastewater treatment decision-making. Researchers used nine years of plant information and tested 19 machine-learning models across 13 operational targets, including flow, energy, pumps and pressure. The work also showed potential for AI-based "soft sensors" to provide additional resilience when physical sensors are unavailable or undergoing maintenance. (ScienceDirect)

Another practical development came this month from Carollo Engineers and its partners, which released an AI & Machine Learning Guidebook for Potable Reuse developed through research funded by the Bureau of Reclamation. The guide is designed to help utilities evaluate, implement and maintain AI/ML applications, and many of its principles can also apply to conventional drinking-water and wastewater operations. (Water Online)

Meanwhile, workforce preparation is becoming just as important as technology. Recent industry guidance emphasizes the idea of the "augmented operator"—using AI and advanced analytics to help operators interpret growing volumes of SCADA and sensor information while keeping trained professionals responsible for operational decisions. (Water Online)

There is a temptation with AI to start with technology.

Don't.

Start with the business problem.

If your estimated-read percentage is increasing, start there.

If aging meters are creating revenue risk, start there.

If field crews are overwhelmed with service orders, start there.

If managers spend hours assembling weekly reports, start there.

Then ask:

If the answer is yes, you have identified a worthwhile AI use case.

Never upload confidential utility or customer data to public AI platforms.

Use anonymized information or enterprise-approved AI solutions, and always follow your organization's data-governance, cybersecurity and privacy policies.

Customer information, SCADA information, network configurations, credentials and sensitive infrastructure information require particular care.

The intelligent meter isn't really about the meter.

It is about what the utility does with the information the meter provides.

A utility may have millions of readings.

The competitive advantage comes from turning those readings into decisions:

Which account should we investigate?

Which meter should we replace?

Which neighborhood needs attention?

Where are we potentially losing revenue?

Where should tomorrow's field resources go?

That's where AI begins to move from an interesting technology to a useful management tool.

#UtilityAIPlaybook #AIForUtilities #WaterAI #NonRevenueWater #NRW #SmartMetering #AMI #GIS #AssetManagement #UtilityLeadership #WaterUtilities