An algorithm with machine learning elements was developed. Once a day, it forms an index of popularity for each SKU (product), taking into account the following parameters:
- Seasonal factor for categories (a table was created, where each category of goods for each calendar week is ranked)
- Brand weight (calculated by a complex formula, based on the current stock for each brand for each category and the margin ratio - mark-up)
- Product sizes with the allocation of the key sizes (kernels) for each group of the goods.
A number of other less important parameters were also taken into account. The weight of each factor can be adjusted separately. As a result, according to a complicated formula, the index of popularity of all goods has been updated each night and a product list created for the next day.