# Building Personalized Customer Experiences with LLMs at Bud Financial and Scotts Miracle-Gro ## Overview This case study examines how two distinct companies - Bud Financial in the banking sector and Scotts Miracle-Gro in the gardening industry - are implementing LLMs and AI technologies to create personalized customer experiences. Both companies leverage Google Cloud's infrastructure and AI services to deliver contextual, accurate, and regulated responses to their customers. ## Bud Financial's Implementation ### Business Context - Bud Financial works with banks and fintech companies to help them understand customers through transactional data - Focus on delivering personalized experiences in a highly regulated financial environment - Developed "YS" solution for in-app customer interactions ### Technical Architecture - Models trained from scratch in offline data environment - Infrastructure running on Google Kubernetes Engine (GKE) - Utilizes mix of TPUs, GPUs, and Intel compute-optimized machine types - Implementation focuses on: - Uses Gemini through Vertex AI API suite - Started with Google's JAP Builder (now Vertex Conversation) ### MLOps Considerations - Heavy focus on explainability and transparency - Model training specifically designed for transactional data understanding - Implementation of deterministic journeys for regulatory compliance - Comprehensive testing approach: - User feedback analysis and monitoring ## Scotts Miracle-Gro Implementation ### Business Context - Developed MyScotty AI assistant for gardening advice and product recommendations - Focus on providing personalized gardening experiences - Handles multimodal interactions including image recognition for plants ### Technical Stack - Built entirely on Google Cloud - Key components: ### MLOps Challenges and Solutions ### Product Recommendation System - Unique approach different from traditional e-commerce recommendations - Requires highly specific and contextualized recommendations - Implementation considerations: ### Knowledge Management - Handles large volumes of unstructured data from research and SMEs - Challenges in unstructured data governance - Solutions include: - Focus on accuracy and currency of information ### Search Implementation - Vertex AI Search implementation for natural language queries - Features: ## Production Considerations ### Testing and Validation - Sandbox testing for risk management - Internal validation before public deployment - Continuous monitoring of response accuracy ### Performance Metrics - Focus on user feedback - Trust and engagement metrics - Response relevance tracking - Personalization effectiveness measurement ### Integration and Deployment - Seamless integration with Google Cloud services - Automated scaling and infrastructure management - Regular updates and improvements based on user interaction ## Results and Impact ### Bud Financial - Improved customer trust through personalized interactions - Double response rate for personalized messages - Enhanced regulatory compliance while maintaining personalization ### Scotts Miracle-Gro - Accelerated development of personalized products - Improved customer communication - More relevant and contextual product recommendations - Enhanced gardening advice delivery