Smarter, faster, leaner: Optimizing the end-to-end supply chain

The SCM thesis Smarter, Faster, Leaner: Optimizing the End-to-End Supply Chain was authored by Javiera Arancibia and Fernanda Esparza and supervised by Dr. María Jesús Sáenz ([email protected]) and Dr. Jaime Macías Aguayo ([email protected]). For more information on this research, please contact the thesis supervisors.

A planning bottleneck in the age of agility

A global consumer goods company faced a growing challenge: planning its end-to-end supply chain with speed and flexibility. Its optimization system, based on mixed-integer programming (MIP), took nearly two hours to generate each plan, limiting the ability to react to real-time changes. Demand fluctuations, raw material delays, or last-minute production maintenance all required a faster response than the system could provide.

In an industry where responsiveness can drive competitive advantage, this delay had a cost. The company needed a planning solution that could maintain quality while dramatically reducing execution time

When every second counts

The legacy planning model covered every aspect of the supply chain, from purchasing raw materials to coordinating production across two plants and managing distribution to regional warehouses. This comprehensive approach, while effective in scope, became a bottleneck due to its computational complexity. Each time the planning team updated demand forecasts, which happened twice daily, they faced a system that simply couldn’t keep up. In some cases, the planning team had to stop the program mid-run and work with a sub-optimal solution.

Our project set out to change that. The goal was ambitious yet clear: maintain the integrity and quality of planning outputs while drastically reducing the time needed to generate them. We tackled this challenge using a two-pronged strategy. First, we broke down the planning problem into smaller, independent segments to reduce the computational load. Then, we implemented a hybrid approach that combined Genetic Algorithm (GA) and Particle Swarm Optimization (PSO), two powerful metaheuristic techniques capable of navigating complex problems with a large number of variables.

To make the system adaptable and scalable, we developed a dynamic configurator capable of identifying and responding to three possible supply-demand scenarios. In the simplest scenario, where the plants had sufficient capacity to meet both current demand and inventory targets, the system followed a streamlined path. In more constrained scenarios, when one plant needed to support another, the configurator shifted strategies to optimize cross-plant coordination. Each case activated a different combination of algorithms that leveraged the GA capacity to handle large solution spaces, and PSO speed.

A near-optimal plan in a fraction of the time

The result was a flexible, intelligent planning configurator that determines what and how much to produce, when to do it, and where to ship each SKU. Compared to the existing MIP system, our solution reduced the execution time by up to 90%, from nearly two hours to just a few minutes.

In addition to speed, the model also improved key supply chain goals. Production costs were reduced by 2% and distribution costs dropped by 13%. Stockouts during the planning horizon dropped significantly, from 3.7 SKUs per day to almost zero. The model also maintained healthier inventory levels, favoring service continuity over risky lean inventory strategies.

Thanks to its rapid execution, the program can be run in almost real time, giving planners a practical and scalable solution for uncertain conditions. As supply chains continue to face pressure for speed and adaptability, this solution offers a clear advantage: a near-optimal plan, delivered in seconds, ready to meet disruption head-on.

SC
MR