Alibaba Digital Supply Chain: From Digitalization to Intelligent Forecasting

The talk introduces Alibaba's digital supply chain, emphasizing the importance of retail supply chain integration and the need for a full‑chain digital and intelligent solution that supports over 20 business units and millions of consumers.

It breaks down supply‑chain decision challenges into three layers: data (ensuring complete, timely, consistent data), analysis (transforming data into actionable insights), and decision (moving from manual to automated, scalable, and reusable decision making).

The presentation describes Alibaba's supply‑chain practice, positioning the supply‑chain department as a backend function that drives cost and service quality, and outlines the evolution from basic to integrated to intelligent supply chains, highlighting capabilities such as visibility, perception, and self‑adjustment.

Digitalization solutions are presented in three steps: building a data middle‑platform for high‑quality data, creating supply‑chain visualizations through linked business metrics, and enabling data‑driven analysis for demand forecasting, inventory turnover, and plan accuracy.

Intelligent solutions are organized into three layers: algorithm capabilities (forecasting, replenishment, routing, pricing), engineering platforms (algorithm and simulation platforms), and business systems (prediction, planning, fulfillment). Several case studies illustrate these layers in action.

The evolution of forecasting technology is traced from statistical methods (EMA, ARIMA) to machine‑learning models (LGB, random forest) and finally to deep‑learning models (TFT, DeepAR), culminating in Alibaba's proprietary Falcon system, which offers higher accuracy and stability for large‑scale time‑series predictions.

Alibaba DChain Forecast, a SaaS product built on Falcon, provides an end‑to‑end, no‑code forecasting service for retail and other industries, featuring high accuracy, ease of use, and API/SDK integration, with options for private‑cloud deployment.

The session concludes with a Q&A addressing model interpretability and commercial deployment of the technology.