This paper investigates the effectiveness of AI for fault localization using only natural-language bug reports in an industrial setting. It benchmarks traditional machine learning models against fine-tuned transformer models using proprietary data from ABB Robotics, finding that traditional models often outperform transformers in this context.