Speaker "Peter Anthony" Details Back



An ensemble model for lead forecasting

Abstract empowers consumers across America to find and purchase their dream homes by connecting them with real estate professionals who can help them. Each connection begins with a consumer submitting an expression of interest in a particular home, thereby generating a lead. Monetizing the millions of leads generated on every year requires forecasting monthly lead volume in over 30,000 individual real estate markets, most of which are tied to single ZIP codes. To improve forecasting accuracy and optimize the distribution of leads to our professional customers, we developed an ensemble forecasting model that exploits the seasonality of the real estate business. We first use a variety of autoregressive algorithms to generate forecasts for a historical period, compare these forecasts with actual lead counts, and compute a weight for each algorithm based on its accuracy. We then apply these weights to forecasts made for a future period of interest, which is offset from the historical period by one or more seasonal cycles. The ensemble model has demonstrated greater accuracy than the spreadsheet-based model it replaces, and has allowed automation of inventory management for one of the company's largest lead products.


Data Scientist