1) The document discusses lessons learned from a bank's initial chatbot implementation, including that top customer requests differed from traditional contact centers and it was difficult to classify messages among 300 intents. 2) To address these issues, the bank redefined the welcome menu/dialog, prioritized the most common questions, used machine learning to cluster messages and identify topics, and created "skilled bots" for each topic. 3) The improved approach had a router bot identify topics and hand off to skilled bots that each focused on a single topic, with an iterative process to refine the bots.