Dynamic Algorithm for Target Inventory and the Impact on Replenishment Strategy

Abstract

This thesis proposes a novel dynamic replenishment algorithm that minimizes total inventory and maximizes customer experience by avoiding stockouts. The algorithm achieves this through optimally setting the days of coverage for every item in every store on a given day based on a number of features. Traditional approaches rely on the base stock model for calculating cycle stock using the demand forecast and safety stock using the demand variability combined with service levels set by the business. This thesis focuses on non-traditional approaches that do not rely on the base stock model, which makes it difficult to dynamically set the optimal target stock levels.

The proposed solution is an optimization-based heuristic that calculates the optimal days of coverage for a given item-store-date combination by combining the existing static estimate for days of coverage with a number of features designed to capture customer behavior, store characteristics, and item attributes. The optimized days of coverage metric is then combined with Zara’s demand forecast for each item to compute target stock levels. Finally, the new target stock values are run through a simulation to understand the impact on stockouts and total inventory.

The heuristic is compared with the baseline approach that is currently used by Zara. The heuristic overperforms the existing approach by 2.2%. This translates to a 2.2% reduction in total inventory with minimal (<0.1%) impact on stockouts. The heuristic consistently overperforms the baseline approach across a wide range of scenarios in different countries, departments, and time frames. Features such as item importance, price, and model type are highly predictive of future demand and help calculate days of coverage more optimally. This heuristic demonstrates that incorporating additional data and calculating target stock dynamically has the potential to enhance inventory replenishment processes at retailers using non-traditional policies.