Email is far and away the most common initial vector for cybersecurity breaches and this problem only continues to grow. In 2018 the FBI estimates over $12 billion was stolen through business email compromise. Attackers have become ever more sophisticated, shifting from broad campaigns to carefully crafted attacks that bypass traditional email security. This talk focuses on why these attacks are difficult to stop and how advances in AI, specifically in machine learning and natural language processing, can be used to prevent advanced email fraud. We will discuss real examples of attacks, analyze why traditional approaches fail, and illustrate how data science is the best approach to preventing business email compromise, advanced spear phishing, and email account compromise.