Ping Xu has held a variety of optimization and demand forecasting roles at Amazon for nearly 15 years. She first joined the company as a full-time employee in 2005, shortly after earning her PhD in operations research from MIT. In this interview, she talks about her journey from her graduate studies at MIT to leading Amazon’s efforts to better manage its massive supply chain; why she joined Amazon; and why scientists at Amazon have such great opportunities to have impact at scale.
Q. Tell us about your journey to Amazon?
I’ve been with Amazon since 2005. I also interned with the company in 2002. At the time, I was working on my PhD at MIT in something called operations research. This is a field that was practically invented during World War II, when the military had to figure out how to move huge amounts of food and supplies to soldiers. Economists and applied mathematicians got together to design algorithms that would help them understand how to move efficiently all the things the military needed. Today operations research sits at the intersection of applied math, computer science, economics, and statistics.
When I was at MIT, the e-commerce contenders at the time were eBay, Amazon, and Yahoo, and it wasn’t clear who would emerge from the pack. I got introduced to a team at Amazon that's now called Modeling and Optimization, but at the time there wasn’t a job family for science interns. They weren’t sure what to do with me; I wasn’t an engineer and wasn’t an MBA. I ended up working on a project related to toys – how we stock them in the fulfillment centers
When I started work at Amazon there were a handful of scientists; that number has increased substantially in the past several years.
Q. What are your current responsibilities?
I manage two organizations — one is the team that does top-line forecasting for company-wide financial and operational planning. Its forecasts are used by finance for planning, multi-year forecasting, earnings calls, and labor, and transportation capacity planning. It aims to provide an integrated growth outlook for Amazon leaders.
Amazon has an amazing array of complex and interesting problems to solve.
Another team I manage has a reasonably large number of scientists that work on demand forecasting for supply chain. We forecast demand for a few hundred million items a day. Our team attempts to answer questions such as, “I have this new supplier – how much should we order from them?” and “We’re just launching Amazon in a new country. How much do we think they’ll be buying"? These days we use a lot of machine learning and AI to achieve that. It’s our job to understand what we’re going to need in the warehouses over the next weeks and months.
Q. You’ve been with Amazon for some time. What keeps you challenged?
Two things: innovation and empowerment.
Amazon has an amazing array of complex and interesting problems to solve. Ultimately, we want to use science to make economic decisions that have a large degree of uncertainty at scale. Some of these problems are classic (e.g., time series forecasting), but they are complex due to the scale and large variety of use cases we need to cover. Jeff Bezos always champions innovation and experimentation. Those are things very relevant to science. Moreover, it is not just about the quality of an idea but the cycle time it takes for us to validate the idea with data, so the speed of experimentation really matters.
Along with that goes the level of ownership we have in our jobs. I am here to empower my team to really understand our supply chain and enable important business decisions. The problems we solve have a level of urgency and impact.
In short, I am here because I feel challenged and feel empowered to make lasting changes.
Q. What advice do you have for scientists considering opportunities at Amazon?
I’d say there are three great reasons to work here.
One is that I’ve seen a common theme in what scientists look for in their work, and that is their desire to find interesting problems. And we have them here. We’re solving super-interesting problems with statistics and machine learning and economics, and we work on a lot of problems that combine large scale with uncertainty.
Recently, for instance, during COVID-19, we needed to estimate the true underlying demand for products such as toilet paper as they were highly sought after and were out of stock. What is the customer need? What signals do we have? How are customer demand patterns going to change as we transition from COVID-19 lock down to then gradually reopening? We had a recent algorithmic breakthrough using deep learning for time series forecasting. Right now, we are also exploring deep reinforcement learning.
The second thing I’d say is that scientists like to work in a place where they can have impact at scale. The opportunity to have that impact here is huge. Our goal is to always work backward from the customer and solve business problems that make their experience better. You’re solving these things because someone cares and needs a solution today. I think a lot about how to empower a team to push against system and organizational boundaries.
The third thing scientists like is a sense of community. People come here because they want to work with a group of smart people and bounce ideas off each other. I think a lot about how to build an environment where we can exchange ideas freely and borrow strengths from one another.
Q. What do you do to keep your mind refreshed?
I’ve been practicing yoga for 15 years, which really helps me relax and focus. Also, I ride a cargo bike to work. It has two seats and a large rain cover in the back, so that I can pick up my two kids from school (ages 1.5 and 4) in any weather. They certainly keep me busy!
