|
Abstract:
|
Many planning decisions need to be made at a long lead time, with high uncertainty, involving complex interactions across functional areas, but with limited historical data to reference with. Examples include supply parts planning, network infrastructure planning, and data center capacity planning. At Google, we tackle these problems with even higher challenges because of the scale, fast growing business, diversity of our product offering, and constant disruptive technological innovations. In this talk, we will share the learnings we have accumulated over years of practice on long term forecasting to support the Google wide machine and data center planning needs, including how to leverage expert knowledge into statistical forecast, drawing analogies to other things following similar growth trajectory, and the interesting problems of forecasting on forecasts and maintaining hierarchical consistency across planning areas.
|
