For computer algorithms, the challenges of online commerce are math problems: What size box do you need for a group of items? What’s the optimal route for a delivery driver? How do we prevent discrimination in vacation rentals?
But these math problems are also people problems — and algorithmic answers work better when they take into account how people will interact with them. Dennis Zhang, of Washington University in St. Louis’ Olin Business School, is focused on adding that human touch to his research into designing and improving online platforms such as Amazon, Alibaba and Airbnb.
“When we design algorithms, we typically think about the abstract goal we want to achieve,” said Zhang, an associate professor of supply chain, operations and technology as well as marketing (courtesy). “But it’s important to consider people’s reactions to the algorithms and use those reactions to not only optimize for the end goal, but also optimize for people’s experiences in following these algorithms.”
This is a lesson Zhang first learned through his love of video games.
“I have been a gamer throughout my life, and I really enjoy playing computer games,” Zhang said. “To design good computer games, we need not only a good storyline, good graphics and effective AI (artificial intelligence), but also a good experience for the players. We cannot always use the most capable AI in the game because human players need to enjoy winning.
“These early childhood experiences always remind me of the importance of the human component in designing algorithms and systems. This hunch became more of a research stream once I studied economics and operations management in business school.”
Zhang completed his undergraduate studies at the University of California, Los Angeles, where he earned dual bachelor’s degrees in mathematics and in electrical engineering and computer sciences. During his third year of studies, Zhang completed an internship on Wall Street as a quantitative developer.
“That experience on Wall Street sparked my interest in economics and social sciences, in general. This is why I chose to major in graduate school in operations management, which intersects quantitative methods, such as optimization, and social science methods, such as economics,” Zhang said.
In 2016, Zhang graduated from Northwestern University’s Kellogg School of Management, from which he earned doctoral and master’s degrees in managerial economics and operations. That same year, he accepted a faculty position at Olin.
Designing algorithms for the greater good
Zhang said the rise of e-commerce platforms has transformed how traditional businesses operate.
“Think about Uber, Airbnb and Amazon,” he said. “Uber does not own any of its fleet, but it is probably larger than any taxi company on earth. Airbnb is probably larger, in terms of the number of rooms, than any hotel group on earth. And Amazon is obviously larger than any retail chain on earth, probably except Walmart.”
He noted that the concerns of digital platforms are different from those of brick-and-mortar owners, who tend to focus on storing and displaying goods or ensuring that their physical inventory of rooms or cabs is fully occupied.
‘A lot of my research has to do with information algorithms. But it also has a societal idea, which is trying to show how we make the overall welfare of people better — not only the profits of the platform — by designing better algorithms.’
Dennis Zhang
Instead, online platforms are concerned with information design, or how to allocate information. And that’s not just a matter of processing data — it’s about designing algorithms that work with people’s priorities and needs. That theme runs through Zhang’s work.
“A lot of my research has to do with information algorithms,” Zhang said. “But it also has a societal idea, which is trying to show how we make the overall welfare of people better — not only the profits of the platform — by designing better algorithms.”
In 2020, he and his co-authors developed a solution to a packaging problem plaguing online retailers such as Amazon and Alibaba. An important process for these online retail giants is packing customers’ items into boxes to be shipped. These businesses typically use bin packing algorithms to guide workers in selecting and packing boxes. The algorithms aim to minimize the box surface area given a set of goods.
However, the solutions from these algorithms sometimes involve complex item orientations, which can frustrate workers and lead them to choose larger boxes, increasing packing time. Zhang’s team created an algorithm improvement that predicted when workers were likely to deviate from algorithm suggestions and adjusted the box recommendations accordingly.
He said this idea can be adapted to other problems, such as incorporating drivers’ input into optimizing delivery routes.
A 2017 paper that Zhang co-authored in the Harvard Business Review took on an even thornier problem: discrimination on online platforms. Previous studies had shown that Airbnb hosts were less likely to approve potential renters whose names were perceived to be African American.
Zhang and his co-authors showed that by introducing a small amount of new data — in this case, one review — they could significantly alleviate this bias. He said the key was determining what kind of bias hosts were exhibiting. Economics literature defines two types of discrimination: “taste-based,” or the idea that someone doesn’t like a specific group; and “statistical-based,” or the idea that someone associates a group with negative traits.
“If we’re dealing with taste-based discrimination, then the optimal strategy is not to provide you any information regarding the person, so there is nothing to discriminate on,” he said. “In a statistical discrimination world, the solution is to provide more information about this person.
“It turns out that this (Airbnb) discrimination is mostly about statistical discrimination. So, it’s important for the platform to provide more information about the accounts.”
Another stream of Zhang’s research concerns improving the machine-learning algorithms used in business.
“Most of these platforms make decisions about how to display information,” he said. “The best way to do this is to run experiments, and you run them by randomly dividing people into different buckets — for the people in Bucket A you use one information strategy, and for the people in Bucket B, you use another strategy.”
He said a large platform may run tens of thousands of these experiments every day. “In my recent research, I look at how we can do these experiments in a much more efficient way so that we can test more business strategies with a given pool of customers.”
Altogether, Zhang has published nearly two dozen scholarly articles during his first eight years at Olin. His work has led to two Olin Awards, which recognize scholarly research with timely, practical applications. He has also won two Reid Teaching Awards, which Olin graduates bestow on faculty members they see as inspirational, energizing and transformative.
In 2022, he was a co-winner of the Production and Operations Management Society Early Career Research Accomplishments Award. One of the most prestigious honors in the operations management field, the award recognized his contributions to platform operations.
