Titolo della tesi: Constrained Assortment Optimization: an integrated non parametric approach for retailer and consumer satisfaction
Retail assortment optimization is a pivotal activity that intricately interweaves customer satisfaction and business profitability. This essential task ensures that the product mix available to customers is not only desirable but also strategically aligned with market trends, consumer preferences, and operational capacities, thereby maximizing overall revenue and consumer choice satisfaction.
This research embarks on the ambitious journey of crafting advanced optimization models and methods that synergize the retailer's strategic vision with operational execution in assortment planning. The quintessence of our study is a non-parametric consumer choice model constructed on the bedrock of historical sales data, a notable departure from the confines of classical choice models. We meticulously incorporate the dynamics of complementarity and substitution effects to render a model that resonates with real-world consumer behavior.
Our innovative approach unfolds in stages; initially adopting a continuous formulation, seamlessly solved using exact solvers, subsequently graduating to a more nuanced Mixed Integer Linear Programming (MILP) formulation. The latter required navigating through computational intricacies stemming from constraints related to product interdependencies, which were adeptly managed through a 'soft formulation'. A uniquely tailored matheuristic, grounded in a bilevel decomposition approach, was then crafted, capitalizing on the heuristic speed and the reliability of established solvers to face the curse of dimensionality arising with large istances.
Our static model flourishes in its adaptability, capable of defining optimal assortments across diverse temporal windows, thereby embracing the seasonality intrinsic to retail products. A comprehensive framework thus emerges, encapsulating the fundamental tenets of assortment problems such as space allocation and supplier relationships, and at the same time offering retailers a robust, adaptable scaffold to tackle the nuances of product assortment optimization.
To fortify the reliability and versatility of our model, an innovative input data generator was developed. This tool simulates a spectrum of retail scenarios, exploring variables such as shelf space allocation, product numerosity, and inter-product dynamics, ensuring the model's robust applicability and scalability across multifarious retail landscapes.
This thesis stands as a beacon of innovation, bridging theoretical profundity with practical relevance, and charting new territories in assortment optimization. Looking ahead, our model’s evolution envisages the integration of new products beyond historical sales data and a thoughtful consideration of sustainability dimensions, marking exciting avenues for future exploration and enhancement in retail assortment optimization.