An American Water Works Association (AWWA) survey just laid out the problem in stark terms: nearly 70% of utility pros see AI changing everything in the next five years, but a mere 15% feel ready for it. That’s a huge gap. Carollo Engineers is trying to fill it with a new AI guidebook for utilities, showing people how to run pilot projects, fix their underlying data problems, and get their existing teams trained. The guide is full of these kinds of real-world steps. But will a book, no matter how practical, actually get this sector moving?
Key Takeaways
- Using AI for predictive maintenance on water infrastructure can cut operational costs by 10-15% within three years.
- Carollo’s guidebook says to start small with pilot projects that have clear KPIs, not a big-bang rollout.
- Fixing your data quality is non-negotiable. The guide recommends building data governance frameworks so your AI models have good data to work with.
- Cybersecurity is a major hurdle. The guide gives specific protocols for locking down AI systems in critical infrastructure.
- The guidebook’s section on people is key: you have to train your existing staff in basic AI and data science for this to work.
Why an 85% Preparedness Gap Exists
That 85% figure from the AWWA survey, the portion of utility pros who feel unprepared for AI, is an alarm bell. The problem is a lack of actionable pathways. Most utilities, especially smaller municipal ones, are already dealing with aging infrastructure, staff shortages, and razor-thin budgets, so they don’t have the internal expertise to even look at AI vendors, much less deploy a system. The Carollo Engineers’ guidebook tackles this head-on. Instead of talking in abstractions, it shows how to apply predictive analytics to a problem like non-revenue water loss. It explains how a model could flag a likely pipe burst in Atlanta’s Morningside neighborhood days before it happens, turning a costly emergency response into a planned repair. Utilities need to see AI explained in terms of concrete wins like that, not just tech jargon.
Why 60% of AI Projects Fail: Bad Data
There’s a reason up to 60% of AI projects are said to fail: bad data. The problem is even worse in utilities, where operational technology (OT) systems create huge volumes of messy, siloed data. Your SCADA system might be logging flow rates and pressure, but if that data isn’t tied to your GIS and customer systems with good metadata, it’s almost useless for an AI. The Carollo guidebook spends a lot of time on data governance, spelling out exactly why you need solid protocols for data collection, cleaning, and standardization. This is the essential, foundational work. I’ve personally seen an AI project for optimizing wastewater pump schedules fall apart because the energy data was full of sensor errors and gaps. The model just spit out junk. As the guide rightly argues, it’s better to spend money on data infrastructure and data scientists (even contractors) first, because deploying AI on bad data guarantees you’ll get unreliable results.
Cybersecurity: Why 75% of Utility Execs Are Worried
A recent Deloitte survey found that 75% of utility execs see cybersecurity as the biggest hurdle to adopting tech like AI. No surprise there. Utilities are critical infrastructure and constant targets for attack. When you connect an AI to your control systems and operational data, you’re opening up new ways for bad actors to get in. The guidebook deals with this directly, laying out security frameworks and best practices. It covers secure development for the AI models themselves, tight access controls, constant monitoring, and having an incident response plan ready for AI-specific problems. They also bring up explainable AI (XAI), which is a must-have for control systems so that human operators can actually understand why the AI is making a certain decision. Security is mandatory. Any AI tool that weakens your security posture just isn’t an option. The guide makes a smart point about baking AI security into your existing operational technology (OT) security, not just your IT plans. This is the right approach because OT security has to prioritize operational safety and uptime above all else, a very different world from standard IT security.
Only 20% of Workers Feel Trained for AI
Any technology project is really a people project, and this is where many AI plans fall short. A Black & Veatch report found that only 20% of utility workers feel they have the right training to work with AI. It’s a factor that’s too often ignored when new systems are rolled out. The Carollo Engineers’ guide argues for focusing on upskilling your current team instead of just trying to hire new people. That means giving everyone foundational AI literacy, providing deeper training for your data analysts and engineers, and using real change management to get people on board. Think about a plant operator with 25 years of experience managing a water purification system. An AI suggesting chemical dosages could feel like a threat, or it could be an amazing new tool. The key is training that operator to understand what the AI does, what its limits are, and how it can make their job easier. The guide gives real advice on this, like running workshops on data interpretation or letting people use simulated AI dashboards. Its detailed section on workforce development is one of its biggest strengths.
Why “Big Bang” AI Implementations Fail in Utilities
The tech world often loves the idea of a “big bang” AI deployment that transforms the business overnight. For a utility, that approach is just plain dangerous. You’re responsible for essential public services, and any major disruption has immediate public health and economic fallout. The Carollo Engineers’ guidebook argues for the opposite: a phased, iterative rollout. Their recommendation is to start with a small pilot project that has a clear goal and measurable KPIs. For instance, a utility could try using an AI model to cut energy costs at a single pumping station. They could track the savings and operational impact in a controlled way, and only scale it up after it’s proven to work. This method lets the organization learn, fix its data processes, and handle problems without risking the whole system. I’ve seen too many ambitious projects fail because they tried to do everything at once. When public health is involved, a cautious and methodical approach is the only way to go.
Bringing AI into the utility sector is becoming essential for improving efficiency, resilience, and sustainability. But getting it right depends on a methodical plan that puts data integrity, cybersecurity, and workforce training first. The Carollo Engineers’ guide offers a roadmap for this, showing that successful adoption comes from careful, strategic steps, not big, risky jumps. To understand the bigger picture, it’s also worth looking at how issues like AI accountability and platform liability are developing, as these will certainly affect critical infrastructure. The ongoing work on US AI policy is another area that will create new rules utilities must follow.
What specific types of AI applications are most relevant for utilities?
The most common uses are predictive maintenance for infrastructure, cutting energy use in treatment plants, forecasting grid load and stability, finding leaks in water networks, and using chatbots for customer service.
How can smaller utilities with limited budgets begin their AI journey?
Start by picking one high-impact problem, like cutting peak energy use at one facility. To keep costs down, you can use cloud-based AI services with their pre-built models or partner with a university or consultant for a small pilot project.
What are the biggest challenges in integrating AI with existing operational technology (OT) systems?
The main challenges are dealing with older OT systems that don’t have modern APIs, use proprietary protocols, and have built-in security holes. You also have to guarantee real-time performance without interrupting operations. Effectively bridging the IT/OT divide is key because their teams have different priorities, IT focuses on data security while OT focuses on physical safety and uptime, and they need to work together.
How does AI help utilities address climate change challenges?
AI can help by optimizing water and energy use, improving forecasts for extreme weather to make the grid more resilient, integrating renewables more efficiently, and finding ways to cut greenhouse gas emissions during operations.
What kind of return on investment (ROI) can utilities expect from AI implementations?
The ROI comes from lower operating costs (less energy, fewer equipment failures), longer asset life, more reliable service, and better resource management. While it varies, it’s common to see efficiency gains of 10% to 30% in specific areas within two to five years of implementation.