Multi-Task Temporal Fusion Transformer for Joint Sales and Inventory Forecasting in Amazon E-Commerce Supply Chain

2025-11-29

AI Summary

To address stockouts and overstocking caused by the decoupled forecasting of sales and inventory in e-commerce supply chains, this paper proposes the Multi-Task Temporal Fusion Transformer (MT-TFT)—the first framework integrating multi-task learning into the Temporal Fusion Transformer (TFT) architecture for joint time-series modeling of sales and inventory. MT-TFT employs a shared temporal encoder to capture cross-task dynamic dependencies and task-specific decoders to enhance prediction accuracy. Evaluated on large-scale real-world Amazon data, it outperforms single-task TFT baselines: sales forecasting RMSE improves by 6.2% and MAPE by 12.7%; inventory forecasting RMSE improves by 6.4% and MAPE by 12.4%. It also significantly surpasses traditional models (e.g., LSTM, GRU). The core contribution is the design of the first multi-task deep temporal learning architecture tailored for demand–supply co-forecasting, effectively capturing complex cross-variable and long-horizon temporal interactions inherent in supply chain dynamics.

Technology Category

• Planning, Routing, and Scheduling: Optimization of Spatio-temporal Systems
• Machine Learning: Time-Series/Data Streams
• Computer Vision: Multi-modal Vision

Application Category

• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs

Abstract

Efficient inventory management and accurate sales forecasting are critical challenges in large-scale e-commerce platforms such as Amazon, where stockouts and overstocking can lead to substantial financial losses and operational inefficiencies. Traditional single-task forecasting models, which focus solely on sales or inventory, often fail to capture the complex temporal dependencies and cross-task interactions that characterize real-world supply chain dynamics. To address this limitation, this study proposes a Multi-Task Temporal Fusion Transformer (TFT-MTL) framework designed for joint sales and inventory forecasting within the Amazon e-commerce ecosystem. The model integrates heterogeneous data sources, including historical sales records, warehouse inventory levels, pricing, promotions, and event-driven factors such as holidays and Prime Day campaigns, through a unified deep learning architecture. A shared encoder captures long-term temporal patterns, while task-specific decoder heads predict sales volume, inventory turnover, and stockout probability simultaneously. Experiments on large-scale real-world datasets demonstrate that the proposed TFT-MTL model significantly outperforms baseline methods such as LSTM, GRU, and single-task TFT. Compared with the single-task TFT model, the proposed approach achieves a 6.2% reduction in Sales RMSE, a 12.7% decrease in Sales MAPE, a 6.4% reduction in Inventory RMSE, and a 12.4% decrease in Inventory MAPE. These results confirm the model's ability to effectively capture multi-dimensional dependencies across supply chain variables. The proposed framework provides an interpretable, data-driven decision support tool for optimizing Amazon's inventory scheduling and demand planning strategies.

Problem

• Joint sales and inventory forecasting in e-commerce supply chain
• Capturing complex temporal dependencies and cross-task interactions
• Integrating heterogeneous data sources for improved forecasting accuracy

Innovation

• Multi-Task Temporal Fusion Transformer for joint forecasting
• Integrates heterogeneous data sources via unified deep learning
• Shared encoder and task-specific decoders capture dependencies