Stocking hundreds of stores every day… It seems impossible to treat each one individually, with a unique approach to allocation and replenishment. But it only seems impossible. With the right technology, analyzing sales history and historical product availability on the level of each product and store becomes achievable. This, in turn, allows us to answer the question of which products have sold best in a given store in the past. This insight becomes the foundation for achieving the key objective of ensuring each store is stocked with products that reflect its sales potential.
Maximizing sales potential
Analyzing both current and historical data, sales shares across different product categories, micro- and macro-trends, and finally, precise sales forecasting enables retailers to anticipate which products will be most popular among customers. Proper stocking is not only about choosing the right models, colors, and sizes but also precisely defining the quantity of each product that should be sent to every store. The aim is to maximize the sales potential of each product and each store, which translates into greater product availability (reducing out-of-stocks) while also lowering overstock levels. For instance, certain stores perform better with premium products or suits, while others achieve higher results with basic items or shirts. Understanding the specifics of a store’s location and the customers who shop there is essential.
In this complex system, technology becomes a crucial ally. Thanks to Machine Learning and Artificial Intelligence, we can conduct effective data analysis, identify patterns and trends, and even predict the future. In a dynamic market, staying in tune with changes is essential. Regularly trained Machine Learning models automate the process of “learning” on the latest, continually updated data.
Allocation packages in fashion industry
Allocation packages were (and often still are) a basic method for optimizing inventory management in the fashion industry. Ordering products from suppliers in predefined packages allows retailers to efficiently distribute goods from the supplier, through the warehouse, to their stores. However, this approach has its drawbacks. Consumer preferences vary, raising the question of whether allocation based on packages truly builds an optimal assortment in each store.
Allocation packages are often based on fixed sets, which might include, for example, 20 T-shirts with 2 blue ones in size S, 3 in size M, 3 in size L, and 2 in size XL, and the same combination for black T-shirts. However, this approach fails to account for the unique sales dynamics in each store, where demand for specific colors and sizes can vary significantly. Product allocation based on packages means that stores receive the same stock, even though not every store sells the same products with equal success. This results in some stores facing excess in certain colors or sizes, while others experience shortages of popular variants.
The problem lies in the lack of flexibility with package-based allocation. The decision of which products should be included in a predefined package is made by experts, often as early as 12 months before products are delivered to stores due to long lead times. Simplified analysis based on average data about color and size distribution does not reflect real demand in different locations.
Allocation based on packages leads to excess inventory in some stores and shortages in others. Overstock results in low turnover, tied-up capital, and a lack of display space for other products. It requires moving such products to another store or returning them to the warehouse, as well as marking down the goods, which lowers the margin. On the other hand, stock shortages result in lost sales and customer frustration, as well as a loss of brand trust due to the temporary unavailability of specific products in particular colors and sizes. Given the growing competition and increasing consumer expectations, an approach that optimizes logistics but not sales can cause more harm than benefit.
The fashion industry should therefore aim for each store to be stocked in a way that reflects its sales potential. But how can one predict which stores will have higher demand and which will have lower demand for specific products, colors, and sizes? This knowledge is embedded in each store’s sales history. The challenge is skillfully analyzing historical data and drawing conclusions that can be translated into precise product allocation across stores.
The move to AI-Driven allocation
Artificial Intelligence is increasingly supporting the fashion industry in identifying patterns and trends, as well as predicting future events. Thanks to AI, large sets of data from various sources can be processed, enabling brands to adapt better to a changing market. These modern tools allow retailers to respond more swiftly to changes in consumer preferences and plan their collections more effectively.
With AI, it’s possible to implement advanced allocation strategies tailored to the potential of each individual store. Machine Learning models learn automatically from historical data, and the knowledge gained this way is used to accurately predict demand for specific products in each store.
The unique challenges of the fashion industry require specialized Machine Learning solutions. Fashion retail deals with relatively low sales volumes, a high variety of products, frequent changes in collections at least twice a year, the need to accommodate different sizes and colors, generally long supplier lead times, limited display space in stores, and more. This means that Machine Learning models must be tailored specifically to the fashion industry and developed by Data Science experts with deep business knowledge of fashion.
Occubee’s approach to AI-Driven allocation
At Occubee, we focus on the customer and their shopping preferences first, with logistical considerations second. Our Machine Learning models, developed specifically for the fashion industry, are trained on the historical sales data of all products in all stores over the last three years, enabling us to capture general market patterns and trends while identifying local differences and the unique potential of each store. We analyze extensive data sets containing information about sales, weather, historical product unavailability (resulting in lost sales), initial and final pricing, promotions, sales, and product attributes such as size, color, material, style, pattern, and more. This holistic data approach allows us, using an advanced Machine Learning framework, to identify complex relationships between the many factors that impact sales. Models trained in this way enable us to forecast sales of each product in each store, allowing us to assess the sales potential of each store and ensure an optimal offer from the customer’s perspective.
At Occubee, we treat allocation and replenishment as separate processes, which is particularly important for seasonal products available only during a specific season. Sales forecasts based on Machine Learning models enable us to identify each store’s and product’s sales potential. Allocation at Occubee is based on this identified potential, considering various additional factors and rules, such as store section capacity, limited display space, current stock levels, assortment diversity requirements, and ensuring size availability. As a result, picking recommendations are generated, specifying which products, in which sizes and colors, and in what quantities should be delivered to each store. During the season and for basic products in continuous sale, replenishment applies. Occubee offers various strategies, such as ensuring the full size range of a given product in-store, restocking products already in-store first, and only then expanding the collection, among others.
Unlike the traditional approach to allocation, where a predefined schedule introduces new models and colors across the network at the same time, Occubee identifies the start and end of a season and the product lifecycle during the season based on changing sales patterns of seasonal products. As a result, Occubee can suggest allocating a specific model and color to different stores at different times. For example, shorts might be introduced to stores near the coast sooner than in mountainous regions, where the spring-summer season begins later.
Our approach is entirely focused on the customer and their shopping preferences. Occubee ensures optimal stock levels in stores, maximizing sales by aligning the offer with customer needs and ensuring high product availability, taking into account sizes and colors. At the same time, we reduce overstock, positively impacting cash flow and optimizing the use of display space in stores. Such an approach is made possible through the use of advanced AI-based technologies but requires a shift away from traditional allocation packages…
