To be successful, online platforms like Amazon, Uber, or Airbnb depend on individual suppliers’ commitment to providing high-quality services or products for customers.
Bad behavior from contributors may damage a company’s reputation and hurt business. Amazon, for example, has been under scrutiny for fake reviews of products sold by third-party sellers. In 2019, Uber temporarily lost its license to operate in London due to activity involving fraudulent drivers that placed passengers at risk.
It may be difficult for platforms to verify if contributors are doing their best to prevent such problems, so it’s important to create strategies to incentivize good behavior. Peng Sun, the J. B. Fuqua Professor of Decision Sciences at Duke University’s Fuqua School of Business, has always been interested in problems related to managing incentives. He recently worked with colleagues to develop efficient contracts that encourage vendors to strive for excellence.
“The question is: How do you design incentive mechanisms so that people are compelled to do the right thing and, instead of cutting corners to reduce costs, they’re actually putting in the right effort to provide high-quality products or services?” says Sun.
The research—done in collaboration with Yong Liang, an associate professor at Tsinghua University, Runyu Tang, an assistant professor at Xi'an Jiaotong University, and Chong Zhang, an assistant professor at Tilburg University—led to the article “Efficient Resource Allocation Contracts to Reduce Adverse Events,” published online last July by the journal Operations Research.
A moral hazard problem
The situation faced by these platforms can be described as a moral hazard problem, a scenario where one party in a contract may indulge in risky behavior because they won’t have to bear the negative costs of those actions. Platforms have already taken steps to deal with these problems. Uber, for example, may ban drivers who have put passengers at risk and YouTube may refrain from promoting videos from channels that have spread fake news.
"For the moment, these are ad hoc practices that platform companies use. What we try to do is find the best way to implement these kinds of measures systematically,” Sun says. He and his collaborators propose using resource allocation to incentivize good behavior.
The resource in question is the total number of online visits to the platform. A supplier’s income is often proportional to the flow of clients clicking on their products or services. And it’s up to the platform to decide which contributors rank higher on the website and hence are likely to receive the highest volume of clicks.
The role of competition
Considering that there are often multiple suppliers offering similar kinds of products or services on each platform, Sun and his colleagues devised a system where the platform can leverage the competition between them to encourage good behavior.
“Intuitively, if an adverse event happens with one product supplier or service provider, you want to punish them by reducing the click flow that supplier gets,” notes Sun. The proposal may sound simple, but there are many challenges involved in implementing it.
If you reduce the flow of clients to a vendor, you may fail to satisfy all of the customers’ demands. To prevent that, the system described in the paper redistributes those clicks among other vendors that offer similar products.
Another issue is that the distribution of the extra demand may create a perverse incentive for other vendors to not work as hard, Sun notes. That’s why this distribution must account for how well other vendors have behaved in the past.
The paper describes how each vendor gets attributed a type of score that accounts for their whole history of adverse events—which may include a bad review from a customer, failure to deliver a product, or any other indicator of a low-quality service. It’s a single number that accounts for the whole history.
“In the paper, we provide a closed form expression, so just a few lines of computer code that allows the platform to compute those scores, or how much I change the allocation of my clickthrough rates among suppliers,” Sun says. Whenever adverse events happen, it triggers the algorithm to recompute the scores of each supplier and recalculate how the clickthrough flow should be redistributed.
He argues that, despite the mathematical complexity of the research, the system developed by the team is fairly easy to implement. “Our proposals thus provide prescriptive guidance for designing easy-to-implement efficient and incentive-compatible (EIC) contracts that suit the need of practitioners,” the authors point out in the paper.
Sun has several other research projects that investigate how to incentivize people to continuously put in the effort to perform a high-quality service, tell the truth, or do the right thing in general.
The traditional economics literature, Sun says, has a long history of studying incentives. But most models developed more than 15 years ago were single-period models. “In these cases, essentially, we face an uncertain situation and we study how to make sure people do the right thing. But now, with the evolution of mathematical models, computational power, and applications that arise from platforms, you want to make sure that things are managed over time dynamically,” Sun says. “This dynamic control is what I work on.”
He adds that researchers are increasingly interested in tackling problems motivated by online platforms. “Platform economy provides a lot of new perspectives compared to the traditional economy. There are new phenomena and new forms of control that companies can implement and we can help them achieve that.”