Learning brand loyalty from sales data


Picture source: http://blog.anybusiness.com.au/

When a variety of products with different brands sold in store, consumer choice seems to occur hierarchically. If consumers choose a brand first and then a product type this purchase pattern is called the brand-primary process. If consumers choose the type first and then a brand, it’s called the type-primary process

Researcher from Daegu Gyeongbuk Institute of Science and Technology (DGIST) employ the nested multinomial logit model for the hierarchical choice. The expectation- maximization algorithm is applied to estimate the primary demand, while treating the observed sales data as an incomplete observation of that demand.

The estimation requires only realistic data: observed sales, product availability and market share information.  The hidden choice structure can be revealed by applying the proposed method to both the brand-primary model and the type-primary model. The one with the higher likelihood is chosen as the demand structure. by modeling the demand hierarchy accurately, retailers can have its assortment plan with less diversity by using the strong substitution within the product group. So the total expected revenue decreases with less diversity, but the inventory cost decreases more.