A Dynamic Price Inference Approach HIVE BOX

HIVE BOX operates 150,000+ parcel lockers, which cover 100+ cities and deliver more than 9,000,000+ parcels daily, in order to promote efficiency of the last mile of logistics in China. However prices that couriers pay for box use of lockers are still depend on artificial rules, and always are the same in terms of district standard. So in this paper we propose a price inference approach characterized by dynamic and personalization, to assist HIVE BOX in decision making for large scale pricing. Based on casual inference, the approach includes a two-stage model to obtain optimal box price, by which first probabilistic demand curve is inferred and next revenue curve of each locker is maximized. And then we further design a win-win launching policy to make the dynamic price more acceptable easier to couriers. By experimental results both on our simulation platform and in practice, we verify that the whole approach (1) outperforms than current standard pricing and improves revenue growth by its personalization; (2) takes timely and correct reaction to market without human intervene by its dynamical evolution; (3) finishes all 150,000+ lockers inference process within 25 minutes such that global pricing becomes possible and (4) to consider win-win launching policy is of great importance when model and algorithm apply in real market.