The Challenge of Keeping Shelves Full
Walmart operates over 4,700 stores in the U.S. alone, with a vast range of products available in every department. Ensuring that the right items are on the shelves at the right time is a massive challenge. For example, during holidays like Thanksgiving, demand for specific items can spike dramatically. If a store runs out of a popular product, customers are left disappointed. On the other hand, overstocking leads to waste, which is both costly and environmentally harmful.
This is where inventory accuracy becomes mission-critical. Retailers like Walmart must balance supply and demand with precision, especially for perishable items like fresh food. A store manager must decide exactly how much of each item to order, ensuring that nothing goes to waste and that shelves remain stocked.
The Limitations of Traditional Forecasting
Walmart has long used backward-looking machine learning models to predict inventory needs. These models analyze historical sales data to forecast demand for the next 30 or 60 days. While effective for steady, recurring, and seasonal trends, they struggle with unexpected events like storms or sudden sales promotions. In such cases, the models may not account for real-time changes in demand, leading to inaccurate predictions.
For instance, if a storm is approaching, the demand for bottled water or emergency supplies may surge. Traditional models might not detect this shift in time, leaving stores unprepared. This is where the limitations of relying solely on historical data become clear.
Introducing Multi-Agent Workflows
To address these challenges, Walmart developed a multi-agent workflow system. This approach moves away from monolithic models and instead uses a sequence of specialized agents that work together to make more accurate decisions.
The first agent gathers real-time data from various sources, such as sales pipelines and external APIs. For example, in the fresh food department, it might pull metadata from BigQuery to understand current inventory levels and sales trends. This context is then passed to the next agent, which synthesizes historical sales data with the new information.
The third agent plays a crucial role by analyzing the combined data and incorporating real-world context. This could include weather patterns, trending events, or local sales promotions. By integrating these factors, the agent can adjust inventory forecasts more accurately, ensuring that stores are prepared for both expected and unexpected demand shifts.
Finally, the fourth agent generates a natural language recommendation, such as increasing stock by a certain percentage or adjusting orders based on the latest insights. This sequential workflow allows Walmart to make more informed, real-time decisions about inventory management.
The Human-in-the-Loop Approach
While AI and machine learning models are powerful tools, they are not infallible. Walmart’s approach includes a human-in-the-loop system, where store managers can review and adjust AI-generated recommendations. This ensures that decisions are not only data-driven but also aligned with local conditions and store-specific needs.
For example, a store manager might notice a local event that could impact demand for certain items. They can manually adjust the AI’s recommendation to account for this, ensuring that inventory levels are optimized for the specific context of their store.
The Benefits of This Approach
This multi-agent workflow system has several key benefits. First, it improves inventory accuracy by incorporating real-time data and external context. Second, it reduces waste by ensuring that stores are not overstocked with items that may not sell. Third, it enhances the customer experience by keeping shelves stocked with the right products at the right time.
For perishable items like fresh food, this system is particularly valuable. It helps store managers make precise decisions about how much to order, reducing the risk of spoilage and waste. This not only saves money but also supports Walmart’s sustainability goals.
Looking Ahead
As AI and machine learning continue to evolve, systems like Walmart’s multi-agent workflow will become even more sophisticated. These models will be able to process even more data in real time, making inventory management more efficient and accurate than ever before.
For retailers of all sizes, the lessons from Walmart’s approach are clear: combining historical data with real-time insights and human oversight can lead to better inventory decisions, improved customer satisfaction, and reduced waste.
{{appel_action}}
Source : InfoQ