Forecasting a global two-sided marketplace poses a unique combination of analytical and technical challenges. At Airbnb, we firmly believe that the best approach to accurately forecasting such a marketplace is to build a causal model of supply and demand interactions. Expressing causal dynamics as a directed graph stands at the heart of Airbnb's Delphi. Delphi is an internal framework that provides a metric graph to express marketplace interaction as well as a graphical interface for scenario building. This requires to overcome analytical and technical challenges. The analytical challenge lies in identifying the relevant drivers and finding an appropriate seasonal model. The technical difficulty lies in estimating the relationships between drivers and providing a fast computational framework to model them. This talk highlights how we have realized a full-stack implementation of a metric-driven approach to forecasting, which took a cross-functional effort of Engineering, Data Science and Finance and has supported product initiatives, strategic decision-making and reporting.
Kai Brusch is the Product Manager for Delphi at Airbnb. He graduated with a MS in Computer Science from the University of Hamburg and a BS in Computer Science from Hamburg University of Applied Sciences. Previously Kai worked on forecasting as a Data Scientist also at Airbnb.
Data Council, PO Box 2087, Wilson, WY 83014, USA - Phone: +1 (415) 800-4938 - Email: community (at) datacouncil.ai