Design Strategy

The team worked through various approaches during this semester. Using King’s Method, initial safety stock numbers for each SKU were created based on demand variability, lead time, and service level. Through demand analysis, the team further looked at demand variability between each SKU to categorize the SKUs into predictable vs unpredictable and stable vs unstable SKUs. This helped to cater the safety stock targets to individual SKUs and their characteristics. Then, the team built a simulation model using a python script that mimics the real-world Coca-Cola's concentrate network with transportation flows and inventory storage at nodes. The team iteratively modified the safety stocks at each node to come up with the least cost scenario while maintaining service level. 

To validate each regression formula used, two metrics were evaluated: the adjusted R-squared values and the p-values of model parameters. Although these were the metrics used to determine predictability of individual SKUS, they were also used to evaluate the effectiveness of model developed to generate pseudo-future demand. By looking at these metrics, the team was able to verify model correlation and statistical significance before using generated values in the model. 

Validation of the simulation model was significantly less straightforward. Firstly, unit tests were executed to verify the correctness of the computational logic. The unit tests included testing a concentrate warehouse to a single syrup plant, a warehouse to multiple plants, two parallel networks on each coast, and then the entire network as a connected supply chain. Unit tests were also conducted to ensure the model worked for individual SKUs before expanding to simulate all of them. Past that, results were also discussed with the client to confirm feasibility of outputs.