TLDR: The research paper introduces the Supply Chain Planning Agent (SCPA) framework, an LLM-based system designed to automate and optimize complex supply chain planning. Deployed at JD.com, SCPA interprets user needs, decomposes tasks, and generates evidence-based plans, significantly improving operational efficiency by reducing data processing time by 40%, increasing plan accuracy by 22%, and boosting stock fulfillment rates by 2-3%. The framework redefines planning as a dynamic, iterative process, integrating data acquisition, plan formulation, execution, diagnosis, and correction to adapt to real-time market conditions.
In the fast-paced world of e-commerce, managing a vast and intricate supply chain is a monumental challenge. Companies like JD.com, one of China’s largest e-commerce platforms, deal with over 10 million unique products across thousands of warehouses and distribution centers, processing tens of millions of orders daily. This scale demands unprecedented efficiency, accuracy, and responsiveness in supply chain planning, which traditionally struggles with dynamic environments, data heterogeneity, and high uncertainty.
A recent research paper, “Leveraging LLM-Based Agents for Intelligent Supply Chain Planning,” introduces a groundbreaking solution: the Supply Chain Planning Agent (SCPA) framework. This innovative system harnesses the power of Large Language Model (LLM)-based agents to transform how supply chain planning is executed, making it more intelligent, adaptive, and efficient.
The Challenge of Modern Supply Chain Planning
Traditional supply chain planning relies on rule-based systems, human expertise, and optimization models. While effective in stable conditions, these methods falter in the face of modern e-commerce complexities. They struggle with diverse and often incomplete data from various sources, volatile demand patterns, and the slow, manual adjustments required when conditions change. This often leads to stockouts, overstocking, and service degradation, highlighting a critical need for more intelligent and adaptive decision-making.
Introducing the Supply Chain Planning Agent (SCPA) Framework
The SCPA framework is designed to address these challenges head-on. It functions as an intelligent assistant that can understand complex domain knowledge, interpret an operator’s needs, break down large tasks into manageable sub-tasks, and leverage or even create new tools to generate evidence-based planning reports. This framework redefines planning as a continuous, closed-loop process involving five key stages:
-
Data Acquisition: Gathering and integrating all relevant operational data, from sales figures to inventory levels, to ensure plans are based on accurate, up-to-date information.
-
Plan Formulation: Generating initial sales and inventory plans, then breaking them down into actionable, department-specific targets.
-
Plan Execution: Aligning procurement with sales forecasts and dynamically adjusting orders to ensure operational feasibility.
-
Plan Diagnosis and Early Warning: Continuously monitoring execution, identifying deviations from planned targets, and pinpointing the root causes of discrepancies.
-
Plan Correction: Analyzing deviations, revising targets as needed, and re-simulating strategies to adapt to market shifts and uncertainties.
This iterative approach allows the system to continuously refine decisions based on real-time feedback, making planning a dynamic and adaptive cycle rather than a static forecast.
How the SCPA Framework Works
The SCPA framework operates through several specialized agents:
-
Intent Classification Agent: Interprets user queries (e.g., “Generate the November sales plan for the Computer Department”) and categorizes them into specific intents like inventory planning, in-stock rate monitoring, or sales and procurement recommendations.
-
Task Orchestration Agent: Translates the classified intent into a structured list of executable sub-tasks. These might include analyzing historical sales data, processing feature data, and formulating the actual plan. This agent also manages an iterative loop, evaluating results after each sub-task and reconfiguring remaining tasks as needed.
-
Task Execution Agents: These agents perform the actual work. The Data Acquisition Agent uses Text-to-SQL techniques to retrieve necessary data from databases. The Data Analysis Agent generates Python code to analyze trends, patterns, and influencing factors, breaking down complex logic into atomic operations for efficiency and interpretability. It can also directly call pre-existing prediction and analysis functions.
-
Plan Correction Agent: Monitors the plan’s execution in real-time, comparing actual performance against planned values. It detects abnormal deviations, identifies potential causes, and recommends timely adjustments to keep the plan aligned with business goals.
This modular and iterative design ensures that the system is robust, adaptable, and capable of handling complex, interdependent supply chain tasks.
Also Read:
- AI-Powered Scenario Generation for Complex Service Environments
- Smart Planning for LLM Agents: Balancing Speed and Expense
Real-World Impact at JD.com
The SCPA framework was deployed within JD.com’s retail supply chain, managing over 10 million SKUs across thousands of locations. The results were significant:
-
Weekly data processing and analysis time was reduced by approximately 40%.
-
The proportion of plans with an accuracy deviation below 5% increased by 22%.
-
Stock fulfillment rates improved by approximately 2–3%.
These empirical validations demonstrate the framework’s ability to enhance human productivity, optimize inventory management, and support more accurate and timely decision-making in high-volume, real-world scenarios. The research highlights the practical value of integrating LLM agents into supply chain management, offering a scalable and adaptable solution for modern large-scale retail operations. For more details, you can read the full paper here.
