BASF Agricultural Solutions is tackling the complexities of its supply chain, which spans 180 production sites and involves over 5,000 distinct value chains. The process of turning active ingredients into final products can take up to two years, and even minor changes in weather or regulations can disrupt operations. To better understand the interplay of local decisions within its global network, BASF has partnered with AlphaEvolve on Google Cloud to develop a digital twin of its supply chain.
Complex Planning Requirements
Creating a single end product requires navigating a bill of materials that can exceed 30 levels deep. Human planners currently make thousands of daily decisions regarding production schedules and safety stock levels. However, the vast scale of the network makes it challenging for planners to see how localized decisions impact the broader supply chain, often leading to excess inventory or production imbalances.
Establishing a Decision Support Model
AlphaEvolve functions as an autonomous coding agent that generates and refines algorithms. The collaboration with Google Cloud and prognostica GmbH aims to enhance decision support rather than replace human input. The initial step involved creating a foundational program that translated demand forecasts into production schedules, which was then refined using three years of historical data.
Evaluating Algorithm Performance
To gauge AlphaEvolve's effectiveness, the team focused on how closely the simulated inventory levels and production decisions aligned with actual historical data. The latest iterations of AlphaEvolve demonstrated more than an 80% improvement in accuracy compared to the initial model, with expectations for further enhancements.
Significant Results Achieved
The refined planning logic yielded immediate improvements, closely mirroring actual historical supply chain performance and significantly reducing error rates. Dr. Goetz Krabbe, BASF's vice president for global supply chain, noted the success of AlphaEvolve in mapping the complex network and replicating human decision-making processes, resulting in a highly accurate digital twin.
Key Features of the Evolved Algorithm
Through extensive experimentation, AlphaEvolve developed a clear and understandable algorithm that encapsulates the operational dynamics of BASF's network. Key features include:
- Production Consolidation: The algorithm optimizes plant time by grouping production amounts effectively.
- Dynamic Safety Stocks: It adjusts safety stock levels to address seasonal demand fluctuations and manage capital costs.
- Network-Wide Coordination: The model identifies dependencies between production tiers, enhancing global asset utilization.
Future Directions
Initial simulations indicate that evolutionary AI can effectively model large-scale, dynamic supply chains. BASF aims to expand this digital twin to encompass its entire global production network, facilitating ongoing simulations, identifying bottlenecks, and optimizing asset use across facilities.
