When Clearcover’s ML team began implementing generative AI solutions in early 2023, we had assumptions about which technical approaches would deliver value. After successfully deploying multiple production GenAI applications, including a claims adjuster assistant and a claim statement collection bot, the reality surprised us. This talk dissects the gap between GenAI theory and practice, sharing concrete examples of approaches we thought we’d need versus what actually worked. I’ll explore why certain “expected” implementation techniques like vector search and fine-tuning may not always be necessary, and how to determine when simpler alternatives suffice. The presentation will cover our journey from proof-of-concept to production, highlighting counterintuitive discoveries around model hosting, prompt engineering, and API integration that ultimately led to scalable solutions within months rather than quarters. Attendees will gain practical frameworks for navigating their own GenAI implementations, learning from our experiences so they can deliver measurable business value while avoiding unnecessary complexities that can delay deployment and adoption.