Examining the state-space complexity of a game is one way to measure its intricacy. This metric counts the possible system configurations or states. In this article, we will delve into the complexities of the “Supply Planning Game.”

When navigating supply chain planning, the goal is to optimize the quantities and timings of outgoing shipments across all nodes and lanes in the network. This requires considering predictions for demand, lead time, incoming supply, production, and inventory status. By representing all these factors in the state space and decision-making processes, our AI can anticipate future probabilistic imbalances between supply and demand. As a result, optimal supply plans can be generated for the entire time horizon of the Supply Chain Network.

 

Let’s illustrate this with an example. Imagine a product (SKU) network made up of 20 nodes and 50 lanes. Even when we modestly assume that we’re only projecting 8-time steps ahead (i.e., a lead time of no more than 8-time steps), we’re met with 1180 continuous variables. For simplicity’s sake, if we were to categorize each variable into one of 10 bins, the State Space Complexity for our Supply Planning Game skyrockets to an astronomical 10^1180.

But remember, this is a basic estimation. Real-world enterprise supply chain planning often throws much more complicated puzzles our way. Drawing a comparison, think of the ancient game of Go. Its 19×19 board has a dimensional state space of black, white, or empty slots. Google DeepMind’s AlphaGo managed to outsmart the world’s best human player in this intricate game. Yet, the state-space complexity of Go is pegged at 3^361, or 10^172. Astonishingly, this is still dwarfed by the 10^1180 complexity of a medium-scale supply planning game!

Scale that up to envision 3000 interconnected SKUs. Each SKU brings with it an 1180-dimensional challenge. Supply Chain Planners are constantly engaged in these epic Probabilistic Supply Chain Planning games.

At Noodle.ai, we harness the power of Generative Probabilistic Planning (GPP) technology to tackle these intricate challenges. Our toolkit includes Generative AI, encompassing Attention-based Graph Neural Nets (GNN), Offline Reinforcement Learning (RL), and Probabilistic PolicySimulation. The outcome? A network-wide optimal, risk-preferential, dynamic, and resilient supply chain plan. We adapt in real-time to changing objectives, such as balancing running out of stock and overstocking while adhering to constraints.

At Noodle.ai, we are experts in supply chain planning and boldly game the system to deliver superior outcomes. Join us to dive into the future of supply chain plan