Understanding and predicting demand through visual characteristics and neural networks

More about the book

Every season, online fashion retailers face critical decisions regarding product orders and quantities, as replenishment is challenging due to short selling seasons and the perishable nature of fashion items. Understanding typical sales patterns and making precise predictions can help prevent costly over- or understocking. Fashion products serve a social function, being visible and capable of communicating identity and belonging, making appearance a key driver in consumer purchase decisions. However, there is limited knowledge about the visual characteristics influencing demand, which is essential for understanding consumer behavior. To address this gap, I assess a product’s design typicality and brand prominence through processing fluency and conspicuous consumption theories, linking these visual traits to sales patterns. Traditional measurement methods often rely on manual evaluations, restricting research to small datasets. I employ advanced neural networks to automate the extraction of visual characteristics, allowing for large-scale analysis of brand prominence. Additionally, I investigate whether incorporating product appearance into sales and return predictions enhances accuracy. By extracting image features from convolutional neural networks, I operationalize appearance alongside design typicality and brand prominence. My findings indicate that while visual characteristics can predict sales, they do not explain or pred