10/2/25 | 4:15pm | E51-145
Ping Xu
VP – Inbound Systems
Amazon
Abstract: In this survey talk, Ping Xu explores Amazon’s journey in applying AI to supply chain management, from early days of manual forecasting to current state-of-the-art AI systems. She frames this evolution through three phases: predictions (past), where Amazon transformed its 17-year-old legacy forecasting system into a unified deep learning model achieving 5-15x accuracy improvements; coordination’s (present), where the company is tackling complex multi-agent problems through Consensus Planning Protocol and reinforcement learning; and autonomous agents (future), where AI systems will independently manage supply chain decisions. Using real examples from Amazon’s retail business, including response to Q4 2022 capacity challenges and recent port strike disruptions, she shares key learnings about the time required for breakthrough innovations, the importance of solving the right problem rather than just improving existing solutions, and the critical balance between big data approaches and small data decision-making in real-world scenarios. Whether working in supply chain or other domains, these insights provide valuable perspective on implementing AI transformations in complex operational environments.
Bio: At Amazon, Ping Xu was the first science intern hired (2002) and the first female scientist hired (2005). She received her PhD in Operations Research from MIT in 2005. Her PhD thesis was motivated by Amazon Supply Chain challenges, and was the first successful collaboration between Amazon and Academia. Ping has 18+ years of experience in predictive modeling and analytics. She was promoted to Director in 2017 and currently leads the Inbound Systems Organization within Amazon’s Supply Chain Optimization Technologies (SCOT) Organization, including Placement, Simulation/Reinforcement Learning, Coordinated Planning, and Grocery Sourcing/Distribution. Prior to that, she led Forecasting, including Topline Forecasting, ASIN Demand Forecasting and Selection. Her work has empowered Amazon to make informed and robust mid/long-range planning decisions during a unique period of systemic shock. This was achieved with an improved guidance of forecast accuracy by 25-35% at the onset of COVID and 980bps in the 2-year period after. Her work has also enabled better supply chain decisions by spearheading Machine Learning (ML) innovations through the application of Deep Learning (DL), Natural Language Processing (NLP), and Reinforcement Learning (RL). Additionally, Ping runs the annual Consumer Science Summit (CSS) and enjoys contributing to the community through mentoring and panel discussions.