EventCast: Hybrid Demand Forecasting in E-Commerce with LLM-Based Event Knowledge

TL;DR

EventCast tackles the challenge of non-stationary, event-driven demand in e-commerce by decoupling semantic reasoning about future events from numerical forecasting. It uses an LLM solely for generating interpretable textual summaries of upcoming campaigns, holidays, and incentives from unstructured business data, which are then embedded and fused with historical demand signals in a dual-tower forecasting architecture. The approach achieves consistent improvements over state-of-the-art baselines, particularly during high-variance event periods, and has been deployed in production across multiple countries with weekly retraining. By providing interpretable, event-aware forecasts and reducing reliance on heavy LLM embeddings or fine-tuning, EventCast offers a practical, scalable solution for industrial time-series forecasting in dynamic e-commerce environments.

Abstract

Demand forecasting is a cornerstone of e-commerce operations, directly impacting inventory planning and fulfillment scheduling. However, existing forecasting systems often fail during high-impact periods such as flash sales, holiday campaigns, and sudden policy interventions, where demand patterns shift abruptly and unpredictably. In this paper, we introduce EventCast, a modular forecasting framework that integrates future event knowledge into time-series prediction. Unlike prior approaches that ignore future interventions or directly use large language models (LLMs) for numerical forecasting, EventCast leverages LLMs solely for event-driven reasoning. Unstructured business data, which covers campaigns, holiday schedules, and seller incentives, from existing operational databases, is processed by an LLM that converts it into interpretable textual summaries leveraging world knowledge for cultural nuances and novel event combinations. These summaries are fused with historical demand features within a dual-tower architecture, enabling accurate, explainable, and scalable forecasts. Deployed on real-world e-commerce scenarios spanning 4 countries of 160 regions over 10 months, EventCast achieves up to 86.9% and 97.7% improvement on MAE and MSE compared to the variant without event knowledge, and reduces MAE by up to 57.0% and MSE by 83.3% versus the best industrial baseline during event-driven periods. EventCast has deployed into real-world industrial pipelines since March 2025, offering a practical solution for improving operational decision-making in dynamic e-commerce environments.

Figures (5)

• Figure 1: The demand volume affected by sales promotions and religious holiday. Reasoning examples are shown. During Ramadan, demand surges sharply in the days leading up to Hari Raya Puasa as it is in a major promotional period, followed by a steep drop as people shift their focus to religious events, while businesses and logistics pause operations.
• Figure 2: Pipeline overview. (1) An Event Expert Database stores business data with curated prompts. (2) An LLM Agent then acts as a domain-aware semantic reasoning agent, interpreting the business operations and converting them into interpretable textual summaries. (3) Summaries are fused with historical signals. (4) The Forecasting Model combines these embedded features for predictions.
• Figure 3: Illustration of feature alignment and fusion.
• Figure 4: The visualization of normalized prediction results. The highlighted area corresponds to a major holiday sales event. We present predictions from our method (in blue), the best baseline (in orange), and the ground truth (in green).
• Figure 5: The effect of $\lambda$ of Eq. \ref{['eq:sum']} on demand forecasting.