RHI Magnesita is a global Mining and Manufacturing Company specializing in refractory products and high-temperature insulating ceramics. This case study explores their journey in implementing their first AI agent deployment, focusing on transforming their customer service operations across North America. ## Background and Challenge The company faced significant challenges in their customer service operations: * $3 million in annual losses attributed to human errors in order processing, invoicing, and accounts receivable * Lack of standardized training processes for Customer Service Representatives (CSRs) * Complex workflow requiring CSRs to access multiple systems and documents * High training costs (approximately $50,000 per training session for 33 people) * Inefficient knowledge retention from traditional training methods * Operations spread across multiple countries (US, Canada, Mexico) ## AI Solution Implementation Working with IT-Tomatic, they developed an Industrial Virtual Advisor AI agent. The implementation process revealed several important aspects of deploying LLMs in production: ### Data Preparation and Training * Collected approximately 300 real CSR questions to train the model * Gathered existing training materials and validated their current relevance * Conducted video interviews to capture current processes * Created new standardized training documentation * Faced challenges in data sharing between systems due to security restrictions * Found workarounds using Microsoft Teams for secure data sharing ### Deployment Strategy * Started with a focused proof of concept (one customer, one order type) * Resisted temptation to expand scope during initial implementation * Emphasized clear messaging about AI's role in augmenting rather than replacing jobs * Implemented in phases to manage expectations and allow for proper training ### Technical Integration * Integration with SAP R3 ERP system * Consolidated access to multiple data sources: * Master price lists * Customer requirements * Transportation information * Payment terms * Special invoicing instructions * Order history * Inventory data ### Model Capabilities The AI agent demonstrates several key functionalities: * Provides detailed order history analysis * Validates order quantities against historical patterns * Offers comprehensive customer requirement information * Performs invoice accuracy checks * Assists with inventory management * Supports order planning and customer management ## Change Management and Adoption The implementation team faced several challenges and learning opportunities: ### Cultural Adaptation * Initial skepticism and misconceptions about AI capabilities * Strong emphasis on messaging that AI augments rather than replaces jobs * Particularly successful adoption by the Mexico-based team * Focus on career development opportunities for employees ### Training and Support * Required strong project management from the vendor team * Needed continuous coaching and motivation * Importance of setting realistic expectations during initial deployment * Managing the transition from excitement to potential disappointment during early stages ## Results and Impact The implementation has shown several positive outcomes: ### Operational Improvements * Reduced time for data analysis and information gathering * Standardized processes across different regions * Better error catching and prevention * Enhanced training efficiency for new CSRs ### Role Evolution * Transformation of entry-level CSR positions into hybrid roles * Integration of supply chain management, sales, and inventory analysis capabilities * Improved career development opportunities * Better support for technical decision-making ### Future Potential * Platform for digital transformation initiatives * Possibility for expansion to other business areas * Potential for global standardization of processes * Support for cross-border management transfers * Real-time KPI tracking capabilities ## Technical Lessons Learned Several key technical insights emerged from the implementation: * Importance of proper data preparation and validation * Need for robust testing of model outputs * Value of real user feedback in prompt engineering * Significance of security considerations in data sharing * Benefits of starting with a focused use case * Importance of continuous model training and refinement ## Future Directions RHI Magnesita plans to expand the AI agent implementation: * Integration with upcoming SAP system upgrade * Extension to supply chain management software * Implementation in transportation management * Support for sales and production forecasting * Development of specialized agents for different business functions * Potential creation of executive-level AI assistants The case study demonstrates the importance of careful planning, clear communication, and focused implementation when deploying LLMs in production environments. It also highlights how AI can transform traditional roles while improving operational efficiency and reducing errors.