Boosting Delivery Time Predictions With Machine Learning
Wayfair is a major internet retailer for furniture and home-goods. Displaying accurate information to our customers while they browse the site increases customer satisfaction and revenue. In this presentation, I will show the recent results of Wayfair’s new delivery time prediction engine, which leverages gradient boosted trees to cut the average contingency between promise and actual delivery by half, significantly increasing conversion rates. I will touch on some of the subtleties of the underlying business problem, such as the speed vs reliability trade-off. The main discussion will focus on the major aspects of the chosen machine learning solution and how it handles the various challenges of this highly dynamic problem. I will conclude with a short discussion on the technical challenges of serving the results of our model at scale to millions of customers.