Neural machine translation (NMT) brings new quality standards in machine translation, and new ways of working for both Language Service Providers as well as multinational companies. Since quality is the major determinant of success, companies need to ensure that their NMT models perform at the required quality standards by improving their NMT model's performance constantly. To do this, the NMT model's performance needs to be monitored. Traditionally, this entails human evaluation performed by post-editors who score the machine translation output according to specific evaluation criteria. Recently, computer linguists have proposed automatic evaluation metrics. This thesis empirically analyses 6 different automatic evaluation metrics to investigate which automatic metric correlates best with human evaluation for the English-German model, which is part of Roche Diagnostics' NMT system DIAlingo, and how the inclusion of NMT impacts the efficiency of translation processes within Roche Diagnostics International Ldt.