The interest of LSP and companies in neural machine translation solutions is growing worldwide and in Switzerland. One large Swiss insurance company that is turning its attention towards MT is la Mobilière. In this master thesis, we have trained a NMT system to try and find out whether a trained system can provide a superior output compared to a generic system, DeepL. This was achieved using automatic metrics (BLEU and TER) and human evaluations. Our results show that a trained system can indeed bring considerable advantages compared to DeepL, especially with technical texts. Finally, we carried out semi-structured interviews with the translators who took part in the evaluation task to know their opinions and stances on machine translation and post-editing.