Use an email phishing engine that takes into account American vernacular as well as spelling and grammar.
Foreign phishers, of which most seem to be, cannot master the idiosyncrasies of American vernacular and they even struggle with basic grammar and spelling. Since they are usually posing as legitimate American companies or professional individuals, create a machine learning engine that understands American vernacular then sanity check emails for violations of the learned rules of the vernacular. Microsoft likely already has a usable machine learning capaability already (bot Tay for example). Time to stop playing with that technology and start using it!