Topic: Quantitative approaches to reduce online return behaviour

The rise of online shopping has brought along with it the issue of high return rates, presenting a significant challenge for e-commerce businesses. Online return behaviour, often driven by factors such as the inability to try on items, misrepresentation of products, or discrepancies in size and fit, leads to increased operational costs for retailers who have to process and restock the returned items. Furthermore, returns negatively impact the environment, as the process of handling and transporting goods back and forth significantly increases the carbon footprint. This return culture is particularly prevalent in the fashion industry, where the return rates are higher than in other sectors. Consequently, understanding and managing online return behaviour has become crucial for retailers aiming to improve their profitability, customer satisfaction, and sustainability.

This dissertation adopts a quantitative approach to product returns. The aim is to explore methods for predicting returns based on product, customer, and order information. Furthermore, the objective is to investigate to what extent practical actions can be linked to these findings, resulting in real-world reductions in return behaviour.

Topic: Modelling size distributions in apparel retail

Estimating the precise quantity and distribution of sizes to procure and distribute as an apparel retailer is a complex task. An inaccurate estimation can result in an excess of certain sizes and a scarcity of others, leading to missed sales opportunities and reduced customer satisfaction. Excessive inventory, a common consequence of overestimation, results in unsold merchandise that frequently ends up in landfills or incinerators. This problem is exacerbated by traditional retailers adopting strategies that aim to maintain a comprehensive range of sizes across all stores. While this strategy reduces lost sales, it inherently leads to surplus inventory.

The goal of this dissertation is to develop new methods to determine the sizes that will be purchased. This task needs to take into account not only better forecasting of true unconstrained demand but also the operational realities of a typical retailer.

As part of this study, realistic data will be generated with the capability to vary select parameters such as the size of the store network, uncertainty of demand, and the composition of the product assortment. This approach will allow for the derivation of general rules about best practices for size management of apparel retailers. If desired, the methods can also be validated using real data.

Topic: Advancing Operational Research through Large Language Models (LLM): An Empirical Business Case Study

This dissertation explores the intersection of operational research and large language models (LLM). The study should is grounded in a real-world business setting, and aims to create an innovative application that contributes to both academic knowledge and practical business operations.

The student(s) must investigate and illustrate the potential of this integration of traditional techniques such as simulation and optimization with novel LLM models. Specifically, the dissertation is likely to focus on two key benefits offered by LLMs (but is not limited to these aspects):

1. The ability to create structured data from unstructured inputs, broadening the range of possible applications.

2. The ability to interpret complex mathematical results, and provide them in a more user-friendly format for the end user. Especially when it comes to sensitivity analysis and the finetuning of models this could be a significant advantage.

It is possible that specific candidate companies and cases will be communicated at a later date, but the responsibility of finding a suitable company and case ultimately lies with the student.

Topic: Re-envisioning Fashion Supply Chain Optimization: Incorporating Environmental Objectives alongside Profit Maximization

Over the years, supply chain optimization has primarily focussed on maximizing profits, employing mathematical models to achieve optimal operational efficiency and cost-effectiveness. The goal of this dissertation is to investigate how these models can be adjusted to account for environmental objectives, as well as profit maximization objectives.

Environmental objectives are often reported in sustainability reports (for example by companies such as Inditex), but the majority of reported parameters relate to the sourcing of materials, rather than the efficiency of operations. The goal of this dissertation is to investigate how methods aimed at improving operational efficiency can also be integrated into these methods.

The dissertation's empirical component involves conducting a computational experiment utilizing real or realistically simulated retail data. The purpose of this experiment is to unveil the potential disparity, if any, between optimizing for conventional economic objectives and for environmentally-centered ones. These findings will not only enrich the theoretical understanding of the integration of environmental objectives into supply chain optimization but may also stimulate industry-wide adoption of eco-conscious business practices.

Topic: Enhancing Sensitivity Analysis in Supply Chain Optimization Through Generative AI: An Investigation into Quality and Efficiency

Sensitivity analysis, while being an indispensable tool in supply chain optimization, often faces neglect due to its laborious nature. The potential of Generative Artificial Intelligence (AI) as a mechanism to streamline and enhance the sensitivity analysis process holds significant promise.

The dissertation will investigate this possibility, aiming to uncover how a generative AI can be instrumental in improving the efficiency and quality of sensitivity analysis. To this end one or more traditional supply chain optimization problems are to be selected and computational experiments are conducted.

A key challenge is to develop a way of measuring the quality and efficiency of the sensitivity analysis as it is performed in an automated fashion, and comparing this to relevant benchmarks. The objective of the dissertation is to contribute significant knowledge, exceeding the anecdotal evidence provided by a single prototype.

By better understanding the implications of employing generative AI in supply chain optimization, the work could offer academics and industry practitioners valuable insight into how they might improve decision-making processes, and achieve better outcomes in terms of cost, time, and resource efficiency.

Topic: A Bayesian approach to inventory management in (fashion) retail

Retail inventory management is about placing the right quantity of inventory in the correct location. Where traditional retail can focus on replenishment and economic order quantity style models, this is not the case for fashion retailers. These retailers are faced with novel products every season, often without the ability to order additional inventory within seasons.

Current strategies often rely on excessive inventory buffers, causing significant waste that has a high environmental impact. While the movement to steer away from fast fashion is growing, a field of tension is likely to remain as long as companies operate in free market conditions. Hence, improvements to company operations are one avenue that must be explored to increase environmental and economic efficiency of fashion companies.

The objective of this dissertation is to investigate the application of Bayesian statistics in this context. Specifically the degree to which predictions can be improved by making use of causal models, relying on underlying cause-effect relationships that are assumed by product experts. Moreover, the additional effect of working with Bayesian updating once sales are observed is to be investigated.

The dissertation works using a real or realistically generated dataset and compares the real-world forecasting performance of multiple models. Real-world performance implies that the degree to which decisions based on different models would be different and structurally better because of the use of such models. ion