Using Machine Learning To Estimate Customers’ Valuations In A Single Product Pricing Problem

Official URL: https://risc01.sabanciuniv.edu/record=b3205828

Estimating customers’ valuations is a goal in retail to learn the customers’ willingness to pay for a product so that the seller can set better price levels and in turn increase revenue. In a single product setting, the strategy determines the optimal price level that generates the highest revenue (reward). A key criterion that affects these prices is the customer profiles in the market. In e-commerce businesses, there is big data consisting of the customers’ sales information. Analyzing and processing this data will provide further insight into the market and the customer profiles, and therefore is promising to contribute to increasing sales revenues. This thesis aims to estimate the customers’ valuations and find the optimal price that brings the highest revenue by applying machine learning algorithms to customer sales information. Throughout this thesis, a range of algorithms has been developed to estimate customer valuations, considering various market structures and the level of prior knowledge available. These market structures encompass parameters such as the number of segments, the segment distributions, and their order, and scenarios such as having deterministic or probabilistic valuations. In the offline setting, where a full dataset is available, the algorithms are designed to obtain initial estimates of valuations, including their lower and upper bounds when they are probabilistic, the distribution of segments, the order of these distributions, and the number of segments in the market. Furthermore, algorithms have been developed for the online setting, where sales information is gathered gradually. The primary objective in this scenario is to increase the seller’s revenue through ongoing analysis and decision-making. The results obtained from the proposed algorithms and methods indicate their success in estimating the market structure and achieving high revenues, which closely align with the expected revenue. These findings highlight the effectiveness and efficiency of the developed algorithms in estimating customer valuations and optimizing revenue for the seller. Overall, the research demonstrates the value of the developed algorithms and methods in estimating crucial parameters, optimizing revenue, and achieving revenue levels that are in line with expected revenue values which we expect to achieve based on the true values of the market structure. These findings have significant implications for businesses operating in different market structures, enabling them to make informed pricing decisions and increase their overall revenue potential.