Programmatic Shelf: elevating customer experience through AI-optimized product placement and intelligent shelf organization

This thesis work is based on an internship project held in Procter & Gamble company in which we investigate the utility of interpretable machine learning models, focusing on the versatile applicability of multiple linear regression in addressing a real-world task. In the essay we explore the programmatic shelf problem, which consists of finding the optimal disposition of products inside shelves, with the goal of maximizing the customer experience and increasing the efficiency of the shelf. In the essay we aim to explore how simple models such multi-linear regression may understand and describe customer behavior and sales trends related to the disposition of products within shelves into the Italian consumer goods market, comparing the results obtained with the ones obtained using statistical approaches. In this project, a state-of-the-art deep learning model is first used for the task of object recognition, with the goal of extracting the placement of products over shelves from a set of photographs; Then, from those data we build a multi-linear regression model with the goal of answering business questions provided by the stakeholders of the project; Finally, we describe the results obtained and the advantages of the machine learning approach versus standard statistical approaches.