An energy provider adopts generative AI to produce field service reports in a flash

To feed the AI, Orange Business AI experts gathered data from 30,000 field service reports, supplemented by the energy provider’s customer service notes and operating procedures. The next step was essential data curation: isolating the most relevant information and grounding the model in an industry glossary. The team refined the prompt engineering (how instructions are phrased for the AI) and adjusted the model’s temperature (the parameter that controls output variability), enabling precise calibration of the LLM. This tuning delivers reliable answers while limiting “hallucinations,” that is, instances where the AI generates plausible but inaccurate information.

Thanks to a continuous improvement loop built on regular check-ins and feedback between the partners, the quality of the results steadily improved. “Beyond the purely technical aspects, change management plays a key role,” notes Pierre Smith. “Dialogue with technician teams and transparency about our AI-training approach helped bring employees on board and ease adoption of the LLM.”

The LLM integrates with the technician’s line-of-business application: the technician enters field data (worksite location, time on site, equipment inspected, etc.), and the AI then structures a client-ready report in real time. To do this, it draws on data from existing reports and on maintenance procedures specific to each piece of equipment. These reports are then sent to team leads, who consolidate them into comprehensive activity reports automatically generated by the LLM. Each customer site thus receives an annual report summarizing the service interventions performed and offering recommendations to ensure equipment upkeep. Simple on the surface, this AI conceals real complexity: the documents must be grounded in verifiable data, follow a logical structure, and be tailored to each customer. They also undergo systematic human validation. “For the proof of concept (PoC) of this tool as well as its scaling, we worked with LightOn’s Alfred LLM hosted on Cloud Avenue,” says Pierre Smith. “This Orange Business hosting solution ensures the security and sovereignty of sensitive data through French-based hosting.”