Product Alert System: identify and respond to fast and slow mover items in luxury retail

Gucci is an Italian luxury fashion house, known for excellent quality of Made in Italy artisanship but also for its constant strive for innovation in different fields, including the digital and technological ones. This shouldn’t come as a surprise: many luxury brands try to embrace emerging technologies to become data driven. Data can be used to solve real company’s problems, ranging from improving business processes to understand customer preferences and needs. Product Alert System (PAS), in this respect, addresses the need of knowing how products behave in different locations. The tool is developed together with Supply Chain and Merchandising team with the focus to support them in their decision-making process to especially optimize item-location assortment. In particular, the system is able to trigger “fast mover” and “slow mover” items, along with the occurrence of an out-of-stock event. Products not selling at a quick rate, or not selling at all, run the risk of stagnating on the shelves or in store warehouses causing inefficiencies in logistics and replenishment processes. If reported in time, they could benefit from a better exposure, targeted marketing campaigns or relocation in a different store. Fast mover items, on the contrary, could result in low on-shelf availability or in an extreme case, in an out-of-stock event. These types of events can have a negative impact on the customers' experience, costing Gucci revenues and damaging brand's reputation. In the Proof of Concept step, where feasibility of technical constrains and business requirements is verified, the scope of analysis is limited only to handbags items, USA region and physical stores. PAS is based on an algorithm that weekly processes numerical features and returns two lists, for fast mover and slow mover items respectively, of item – location combinations ranked by a relevance score. The attributes describe and combine together sales, stock and sales forecast trend, sales frequency and information about target stock level and sales price. They are subsequently standardized to make them properly comparable. Partial scores are assigned to each feature according to the Gaussian percentile each standardized value belongs to. The final score is the weighted sum of the partial ones. The higher the score, the more likely the item is a fast mover. The opposite applies for slow mover. PAS project aims at not only providing an informative view of how products perform across locations, but also suggesting meaningful and specific actions to business to tackle inefficiencies and to exploit promising performances. Example of actions can be boost ads or enhance the display in a store for a specific product or change status of an article in the merchandising catalogue. In this regard, results are displayed through an interactive webapp, developed using Dash, that communicates with the algorithm results. Stakeholders and end users are enabled to interact with plots provided, performing filtering on queries, validate the criticality of the item - location combinations shown. The power of the algorithm is tested and subject to validation by merchandisers involved in the process on the ability both to rank useful alerts and to suggest relevant actions to the users. If Minimum Viable Product step, which consists of creating the first prototype of the model, is successful, PAS can be scaled up expanding the scope to other countries or to additional items.