# AI strategy at utilities: value before technology > What separates a real AI strategy from an automation agenda, and what it actually consists of. - **Author:** Dominic Asche - **Date:** 04 / 2026 - **Reading time:** 4 min - **URL:** https://fse-group.de/en/thinktank/ki-strategie-im-evu-wertschoepfung-vor-technologie According to Digital@EVU 2026, 58% of energy suppliers are planning an AI strategy or already implementing one. But ask what that strategy actually contains and you find that much of it is an automation agenda with a new name. An email that automatically becomes a ticket is not an AI use case. So what is one, and what does an AI strategy that holds up consist of? ### The framework According to Digital@EVU 2026, 58% of energy suppliers are planning an AI strategy or already implementing one. That sounds like momentum. But ask what exactly that strategy contains — which processes, which responsibilities, which guardrails — and you quickly notice that much of what is called an AI strategy is an automation agenda with a new name. That is not an accusation, it is a problem of orientation. AI and automation are not the same thing. Anyone who does not distinguish between them builds a strategy on the wrong foundation. ### Position: what AI is, and what it is not An email that automatically becomes a ticket is not an AI use case. A chatbot that routes questions by keyword is not an AI use case. Those are rule-based automations. Useful, but structurally different from what an AI strategy stands for. AI creates value where systems learn from data, recognise patterns, understand context and adapt to new situations without every rule having been programmed explicitly beforehand. The difference is not academic. It decides whether an investment in AI produces real value or merely digitalises processes that were already bad. The right strategic question is therefore not “where can we automate?” It is “in which end-to-end processes do we create measurably more value because a system recognises connections that people alone do not see, or do not see fast enough?” ### A practical example: two use cases, one difference An energy supplier wants to take pressure off customer service. Option A: incoming emails are categorised by subject keywords and routed automatically. Lead time falls, effort stays. Option B: a system analyses content, tone, history and customer profile, detects escalation risks, proposes context-specific solutions and learns from every closed case. Lead time falls, quality rises, and the system gets better over time. Option A is automation. Option B is AI. Both have their place. But only option B justifies an AI strategy. On the grid side the picture is similar: prescriptive maintenance that recognises failure patterns in historical operating data and proposes maintenance windows before a fault occurs is AI. A maintenance plan triggered at fixed intervals is process automation. ### The six pillars of an AI strategy at a utility AI strategy at a utility 01 Maturity & Roadmap 02 Value & Use Cases 03 Data foundation Integration & quality 04 Adoption & co-determination 05 Governance & AI steering 06 Law & Compliance #### 01: maturity & roadmap Every strategy starts with an honest assessment: where does the company stand today? Are there already assisted processes, or are you starting from zero? The maturity picture determines what makes sense as a first step and what is still premature. For utilities, level 3 to 4 on a five-level scale is the realistic target: hybrid teams in which people and machines complement each other sensibly. Full autonomy is a special case for narrow, low-risk sub-processes. #### 02: value & use cases Prioritise two to four end-to-end processes that create real value through AI, not through automation. On the customer side: context-sensitive service processes, intelligent channel steering, detection of escalation risks. On the grid side: digital twins, prescriptive maintenance, adaptive grid planning. The test question per use case: does the system learn? Does it understand context? Does it get better with more data? If not, it is not an AI use case. #### 03: data foundation The bottleneck is rarely the model. It is data access, data quality and system integration. According to BDEW, missing data infrastructure is the biggest operational hurdle. Anyone who does not invest here buys a powerful tool for a building site with no materials. #### 04: adoption & co-determination Introducing AI fails less often on technology than on people. Fear of job loss is real. If it is not addressed it produces quiet refusal or active sabotage. The answer is not a communications campaign at the end but early, genuine involvement: the works council in step two, business units in step three, teams down to the smallest unit. The German Federal Ministry of Labour explicitly describes the role of co-determination in AI adoption as relevant in legal and data protection terms. Change management is a success factor, not an accompanying programme. #### 05: governance & AI steering Who decides which AI initiatives are started? Who makes sure all activities are coordinated along the maturity model and do not contradict each other? Without a central steering function — something like an AI harness that orchestrates and controls all AI activity — you get isolated solutions. They do not scale, and nobody has an overview of how they interact. This function does not have to be large. But it has to exist, have decision-making authority and keep the overall picture. #### 06: law & compliance The EU AI Act comes into force in stages: AI literacy obligations since February 2025, full applicability from August 2026. HR and workforce management systems are classified as high-risk. Utilities affected by NIS-2 have to report security incidents and document risk management measures. That is a leadership decision, not purely an IT task. ### In closing: less paper, more readiness to decide The industry is aware of AI. What is missing is clarity about what an AI strategy really is and what separates it from an automation agenda. The answer does not lie in the technology but in the question: where do we as an organisation actually want to grow, and which processes carry us there? Through systems that think, not only execute. Anyone who can answer that clearly has taken the most important step. The rest is sequencing. #### Sources - BDEW / Digital@EVU study 2026 — digitalisation and AI in the energy sector - European Commission: AI literacy FAQs, digital-strategy.ec.europa.eu - German Federal Ministry of Labour: INQA-109 — using AI in the workplace, 2024/2025