Hardly any other topic is currently dominating industry discussions quite as much as artificial intelligence. At every conference, in every strategy paper, in every conversation with software providers: AI is the answer. Exactly which question it answers, however, often remains unclear.
I have been exploring AI use cases in the energy sector for years. And I have noticed a gap: between what AI promises and what it actually delivers in practice. Understanding this gap is the first step towards a meaningful AI strategy.
What AI can do, and what it cannot
AI is not a panacea. It is a tool with specific strengths and clear limitations.
What AI is particularly good at: identifying complex patterns in large amounts of data, making predictions based on historical data, and supporting processes that do not require 100% accuracy.
What AI cannot do: explore new areas of application without human training, execute processes without error, or draw meaningful conclusions without carefully prepared data. Anyone who thinks they can simply ‘switch on’ AI is bound to be disappointed.
Where AI delivers real added value in municipal utilities
In my consulting practice, I see three areas where AI has been proven to work:
Churn prevention: Predictive analytics identifies at-risk customers before they leave, using dozens of factors that no human could possibly keep track of all at once. Without AI, this simply isn’t possible.
Detection of issues and automated response: Incoming customer enquiries, whether via email, chat or form, are automatically detected, classified and routed to the appropriate process. This significantly reduces the workload in customer service. It is important to note that the technology has clear limitations. Complex or emotional issues still require human intervention. AI handles the routine tasks, not the exceptions.
Liquidity planning: AI-powered forecasts of cash flows, bad debts and dunning processes help to identify financial risks at an early stage. This is an underestimated lever, particularly in a market environment characterised by volatile energy prices.
The next level: Agentic AI
What has changed fundamentally in recent years is the shift towards so-called agentic AI: AI that not only analyses but also carries out tasks independently. An agent identifies a customer enquiry, accesses the CRM system, reads the contract, and directly triggers the appropriate process without human intervention.
That sounds tempting. And it is technically possible.
But here lies the key limitation: agentic AI is only effective if it is deeply integrated into existing systems. It must be able to read data, access processes and trigger actions within systems. AI that operates in isolation is nothing more than an expensive gimmick.
In practice, it is exactly this integration that proves to be the biggest hurdle. Data silos, interface issues and legacy system landscapes are not mere technical details. They are strategic challenges.
The real key: data strategy
And that brings us to the crux of the matter.
The best AI model is only as good as the data it works with. Incomplete data, inconsistent quality and missing historical records lead to models that fail in practice. People are then quick to say, “AI doesn’t work for us.” Yet the problem rarely lies with the AI itself.
Before a municipal utility invests in AI, three questions need to be answered: What data do we have? What is the quality of this data? And what data is still missing?
If you can't answer these questions, that's where you should start, not with the AI tool.
The biggest hurdles in practice
In my experience, AI projects in the energy sector fail for the same reasons
A lack of basic understanding of what AI is and what it can realistically achieve. Most AI projects fail because there is a huge gap between expectations and reality.
No data strategy. No fuel, no engine. It’s as simple as that.
Lack of system integration. AI does not deliver any added value when used as a stand-alone solution. It must be an integral part of the process landscape, not just an add-on.
Inappropriate use cases. In any scenario where 100% accuracy is required, AI is simply not the right tool for the job.
Conclusion
AI is neither just hype nor a magic bullet. It is a powerful tool, provided you understand what it is suited for, have the right data, and take system integration into account.
A holistic approach does not start with technology. It starts with an honest assessment: what do we want to achieve, and are we ready for it?