Data-driven AI Techniques for Fashion and Apparel Retailing

Abstract: Digitalization allows companies to develop many new ways of interacting with customers andother stakeholders. These digital interactions typically generate data that can be stored andlater processed for different objectives. Currently, the fashion and apparel industry isundergoing a disruptive transformation due to digitalization, including a rapid increase in thegeneration of data in various parts of the supply chain. While most data may not be storedwith data mining or other analysis in mind, collected data frequently contain very valuableinformation that can be exploited. Analytics, in particular using data-driven AI techniques, istherefore becoming a pervasive tool, used for a large variety of purposes and in manydifferent processes. While the popularity of Artificial Intelligence (AI) as an advanced tool forimproved decision support is increasing, the applications of AI within the fashion and apparelindustry have historically been rather limited.With this in mind, the overall purpose of this thesis is to, after presenting an overview ofresearch on applications of data-driven AI in the fashion and apparel industry, demonstratehow various data sets and AI techniques can be utilized for improved decision support indifferent scenarios.Whilst the thesis first investigates the impact of AI on different parts of the supply chain, theempirical work focuses on fashion and apparel retailing. Here, different AI techniques areexplored in a set of case studies covering several applications in fashion and apparel retailing,thus showing the potential of utilizing data-driven AI for decision support in that domain.One important learning outcome, found from several of the studies, is the need for combiningseveral data sources and techniques in the projects. Another general result is the benefit ofinterpretable models, allowing inspection and analyses of the relationships found. From anapplied perspective, some approaches, like RFM modeling as a pre-step to churn prediction,adding sentiment analysis to short-term sales forecasting, and building a campaign andsimulation engine from historical data, could potentially be used by many retailers.In conclusion, this thesis has, mainly through a set of case studies addressing real-worldproblems and utilizing real-world data sets, demonstrated how data-driven AI techniques cansupport and improve fashion and apparel retailers’ decision-making.