Paper Title
Conversational Analytics for Supply Chain Operations using Gen AI and Agents

Abstract
A supply chain ecosystem is a vast chain of interconnected links with so many moving parts and data changing all the time that getting timely answers is difficult. Pre-built aggregated views can help address this challenge to a fair degree, but it cannot generate insights on the fly based on a particular situation or simulation. This paper explores a conversational analytics framework that involves GenAI and agentic capabilities allowing Supply chain analysts to query the data through natural and plain language. The response is a structured one that explains the context of the data and provides additional insights. This is accomplished through a combination of agents which analyzes the incoming query (or) analytical workload and through delegation to each specific agent, the response is delivered. The following are the five task driven agents that will perform a series of activities. An intent agent that understands what the user is trying to ask. This will then invoke the context agent which will map the natural language to business specific knowledge. This is then fed to a query agent which generates the technical query to generate the response. The response is then validated through a rule based LLM that acts as a diagnostic agent and an insight agent generates further insights and follow-up questions to the previous chain of thought. This framework integrates live feeds, data from ERP system and external risk related information into Snowflake. Evaluation of this approach showed that there was approximately 60% reduction in lead time, 15-20% improvement in demand forecasting, savings of around 10% and task completion and follow up rose to almost 85%. A conversational framework versus traditional BI dashboards has transformed how analysts operate with data and have resulted in improvement in classification accuracy to 92% and response time to 45 seconds as opposed to 78% & 3 minutes respectively. Keywords - Generative AI; Agentic Workflows; Conversational Analytics; Supply Chain Management; Large Language Models; Multi-Agent Systems; Retrieval-Augmented Generation; Snowflake