Forecasting with Big Data at Google

Abstract

The need for reliable forecasts at scale has brought up new challenges and opportunities at Google. From abundant data sources, time series come in both in a finer resolution and with a wider coverage. These features not only improve the quality of the forecasts but also help address the questions we were not able to decades ago. Meanwhile, they also introduce new challenges such as scalability, robustness and consistency. In this talk, I will briefly introduce the use of time series forecasting at Google, touch on some new opportunities and challenges, and some best practices. A case study of different forecast tools will be provided for illustration.