How Prokon Makes Critical Infrastructure Safer with AI

Prokon generates around one percent of Germany’s electricity and is the country’s largest energy cooperative. The company develops, builds, and operates onshore wind turbines in four European countries. The electricity produced is sold on markets and delivered directly to consumers and more than 50,000 cooperative members through its own B2C business.

77 wind farms, 20 different turbine generations, 2,600 technical documents with more than 50,000 pages. When a turbine fails, every minute counts. Technicians often drive to the same turbine multiple times because diagnoses have to be made on-site and the relevant knowledge is not readily accessible. An average of six site visits per fault means: high costs, long downtimes, and physical strain on the team.

Halved site visits, 25 percent faster on-site problem resolution, measurably higher turbine availability. The technical team works more focused, and the workload drops noticeably.

Prokon operates critical infrastructure. Every diagnostic error can cause six-figure damages. If a component is treated incorrectly and a crane dismantling becomes necessary, the costs run into the millions.

The problem is not a lack of competence. The technicians are highly qualified; many have degrees in electrical engineering or mechanical engineering. The problem is complexity. 20 different turbine generations mean 20 different systems with different fault codes, different manufacturers, different manuals. Knowledge is distributed across documents, monitoring data, and the experience of second-level support.

When a fault code appears, technicians must drive hundreds of kilometers, climb older turbines without an elevator, and then decide on-site: Which component is affected? Which measurements are needed? Which equipment do I need? Often a part is missing, often the first diagnosis is incomplete. Then comes the second site visit. And sometimes the third.

Standard AI solutions fail here. They give uncertain answers, cannot perform multi-step analyses, and hallucinate when in doubt. In critical infrastructure, that is not an option.

Prokon wanted to achieve three things:

1. Fewer site visits through better preparation. Every avoided trip saves costs, increases availability, and protects the team.
2. Reliable diagnoses without hallucinations. When the system does not know something, it must communicate that clearly instead of guessing.
3. Scaling across borders. Poland, Finland, and Spain are on the roadmap. The solution must work multilingually without requiring reimplementation each time.

Keyword search gives you a document. A decision tree gives you a script. octonomy gives you a diagnosis — it reasons through the fault the way your best technician would. It takes a fault code, compares it with current monitoring data, searches relevant manuals, and derives step by step which components could be affected.

For Prokon, this means: a technician does not receive a general note like “check the electrical system”, but a structured diagnostic plan. Which voltage needs to be measured at which component? Which components are possible fault sources? Which tools are required?

• Multi-step fault analysis: The system combines information from different sources and thinks in logical steps, similar to an experienced technician who works through and narrows down different possibilities.
• Seamless integration: Connection to Microsoft Dynamics Field Service and monitoring systems was done via APIs. The existing IT infrastructure remained unchanged.
• 96 percent accuracy: Standard LLMs often score below 40 percent on complex technical questions. octonomy achieves 96 percent because the system works in a structured way and only accesses verified sources.
• Compliance without compromise: Made in Germany, GDPR-compliant, EU AI Act-ready. For critical infrastructure, this is a necessity, not a nice-to-have.

Prokon chose the pragmatic path. No months-long project, but a fast test with a clear focus. After the first conversation, it took two weeks until all relevant stakeholders were on board. The feasibility study with measurable results was completed after ten days. The feedback from the technical service exceeded expectations.
One result stands out. The team’s best technicians now spend their time on the cases that actually need their expertise. Fewer unnecessary trips. Less stress. More focus where it counts.

• Integration: Microsoft Dynamics Field Service as the central platform, directly in the workflow of dispatchers and technicians.
• Data sources: Monitoring systems of the wind turbines, 2,600 manufacturer documents, internal process documentation, experience from second-level support.
• Rollout: Start with a test group in the German market, parallel productive test phase, iterative expansion based on feedback.
• Timeline: Feasibility study ten days, first productive use within six weeks.