Optimizing revenue while showing relevant assortments at scale

Scalable real-time assortment optimization has become essential in e-commerce operations due to the need for personalization and availability of a large variety items. While this can be done when there are simplistic choices made, process becomes difficult imposing constraints on collection relevant assortments based insights by store-managers historically well-performing assortments. We design fast flexible algorithms variations binary search that find (approximately) optimal regime. In particular, we revisit problem large-scale under multinomial logit choice model without any assumptions structure feasible speed up comparison steps using advances similarity field information retrieval/machine learning. For an arbitrary assortments, our solution time is sub-linear number simpler case cardinality - linear items (existing methods quadratic or worse). Empirical validations real world dataset (in addition experiments semi-synthetic data Billion Prices several retail transaction datasets) show competitive (3×−100× faster) even ∼105 (10× larger instances than previously studied).