A Better Conversation

This post introduces the Customer Service Engine (CSE) and its application at a large tools manufacturing, wholesale, and retail business. It details the challenges of building scalable, accurate, and consistent AI for customer service, leading to the adoption of a multi-agent architecture to overcome context window constraints and improve query response times. The system leverages an Ontology tailored to the partner's operations and integrates data from SAP, Salesforce, UPS, and ERP systems. Approximately 50 deterministic 'tools' (retrieval mechanisms, similarity search, pricing/availability models, external web-hooks) were developed and incorporated into Palantir AIP for AI agents. The CSE achieved over 90% query resolution within six weeks, with complex cases escalated for human review and a feedback loop for continuous learning. Key impact metrics include faster response times (<10 seconds for resolution actions), reduced repeat queries (75+%), and improved fraud detection.