The present study explores the hypothesis that artificial intelligence has an edge over humans when translating metaphors in financial and economic discourse. The study builds on evidence that many metaphors can be translated literally from one language to another. It analyzes both the professional and the AI-produced translation into Spanish and French of a large sample of linguistic metaphors found in a corpus of English financial and economic reports. Results show that literal translation was the most frequently used technique in both the Spanish and French human translations, but the percentages differ considerably (around 70% and 45%, respectively). Additionally, in Spanish, the metaphorical load was retained (through either a literal rendering or a different metaphor) to a significantly greater extent than in French, which resorted more often to paraphrasing. These marked differences between the target languages point to a higher tolerance of Spanish to foreign metaphors. Importantly, they were not observed in the automated translations. Here, literal translation was also the most common technique (ChatGPT: 48% in Spanish and 42% in French; DeepL: 58% and 49%, respectively). ChatGPT and DeepL produced acceptable translations in a high percentage of the cases (slightly over 80% and around 76%, respectively, in both Spanish and French), but the vast majority of their mistranslations stemmed from literal renderings, which calls for caution. eTranslation (Finance version) achieved low success rates (under 45%). Finally, complex and chained metaphors increase the level of difficulty for machine translation tools, while they seem unproblematic for human translators.