From Forecasts to Decisions: Integrating Ensemble Forecasting and Private LLMs for Global After-Sales Planning
Accurate demand forecasting is central to effective after-sales supply planning, yet it remains especially challenging in global environments characterized by intermittent demand, long tails, and high service-level expectations. In this presentation, we share an industry case study from TVS Motor Company that demonstrates how to integrate ensemble forecasting, prescriptive optimization, and private large language models (LLMs) into a scalable, enterprise-grade planning system.
The solution combines statistical, machine learning, and deep learning forecasts using a horizon-aware, optimization-based weighting scheme to improve accuracy across more than 6,000 spare parts and 90+ countries. Beyond predictive performance, the system addresses a critical adoption gap: translating forecast outputs into actionable insights for planners and executives. To bridge this gap, we deployed an on-premises private LLM that generates role-specific narratives, scorecards, and diagnostics directly from forecast results, without exposing sensitive data outside the enterprise boundary.
The presentation focuses on practical design choices, implementation architecture, and lessons learned from deployment at scale. We highlight measurable improvements in accuracy, reductions in manual reporting effort, and greater alignment between analytics teams and business users. Attendees will gain concrete insights into how LLMs can augment rather than replace optimization-driven planning workflows, and how explainability and security can be preserved in real-world supply chain analytics.
Essential /
Professional /
Leadership
