ABSTRACT
Dynamic price changes in the growing retail industry can be determined by geography, inventory, and competition, among other relevant factors. A Monte Carlo Tree Search (MCTS) model is proposed to determine the optimal retail sales price for each day, with the objective of maximizing profits with limited inventory. The expected units based on the defined price are estimated through a multiple linear regression model, in which random variables are added to make the simulations effective. The experiment is performed in real- time, on three products of a grocery shopping delivery startup, for five days. The results indicate an increase of 12% in net income and 33% increase in units sold within the test days, in comparison to last week. Since, in essence, the proposed MCTS model must be unsupervised in order to be implemented on catalogs larger than a thousand SKUs, criteria and thresholds must be defined to ensure that the output is optimal.
