You are working on a service bot that handles customer support conversations across chat and case intake flows. The bot needs to understand intent, pull out key entities like order numbers or product names, and respond in a way that matches the customer’s issue without escalating unnecessarily. The current system works for common requests but struggles with ambiguous phrasing, multi-intent messages, and domain-specific language that appears in support transcripts and Salesforce Service Cloud case notes.
How would you enhance natural language processing for service bots?