Deep Learning Based Simulation For New Product Demand Estimation

Increasingly, manufacturers and third-party sellers sell their products online at large e-commerce platforms. A good forecast is necessary to avoid resource wastage in form of over-production or underserved demand. Sensing the demand for a new product on e-commerce platforms in the presence of other competitive products requires techniques beyond traditional forecasting methods. Typical e-commerce platforms, being a two-sided marketplace, need to make a trade-off between showing very relevant and trusted products versus exploring new selections. As customers heavily rely upon organic search functionality for product discovery, the demand forecasting for products depends upon the signals from search algorithms. In this paper, we propose an end-to-end Deep Learning (DL) based solution to estimate new product demand. Our design science research (DSR) approach involves the use of search ranking algorithm and its features to simulate the impressions of new products in the presence of existing products and vice versa. This approach helps to estimate demand with help of simulated search funnel metrics to bake in competitive characteristics of other products.