As online marketplaces such as Amazon and Airbnb come to dominate increasingly wide swaths of the American economy, one big question for the platform operators is how they can generate the most revenue for themselves and their vendors. Would Amazon and its partners be best off if the retailing behemoth shared the vast amounts of data it gathers on consumers and their behavior in various sectors, or not?
Chicago Booth’s John R. Birge, Chinese University of Hong Kong’s Hongfan Chen (a graduate of Booth’s PhD Program), Duke’s N. Bora Keskin, and Booth’s Amy Ward write that, “perhaps surprisingly,” a policy such as Amazon’s, in which a platform tells its vendors absolutely nothing, isn’t necessarily terrible—but there may be a more optimal level of data-sharing.
Online platforms make money by collecting commissions from individual sellers, such as homeowners offering their places to travelers on Airbnb or freelancers peddling their services on Upwork. In most cases, sellers set their own rates. A lack of information on consumer demand can mean they fail to set optimal prices, resulting in lost revenue for themselves and the platforms. Meanwhile, the platforms are in a position to compile tons of data on clients over time, including browsing and purchasing histories and personal information that may affect buying choices.
Different platforms follow different models for information sharing with vendors. Airbnb provides estimated earnings information to hosts based on factors including booking data in their area. It also offers price-setting incentives. For example, hosts pay a fee to join the platform’s premium program, which then steers them toward earning a “Plus” classification and higher listing prices. Upwork shares reports on the popularity of different tasks. And then there’s Amazon’s practice of not sharing demand information with vendors—what the researchers refer to as a “do-nothing” policy.
For most platforms, the do-nothing approach wouldn’t be ideal because the lack of transparency means sellers are largely in the dark on price setting, the researchers observe. However, they find there are times when doing nothing might work to both the sellers’ and a platform’s advantage.
In the case of Airbnb, Birge says, a do-nothing approach might work in a smaller city where two hosts in different neighborhoods charge the same high prices, are both booking rooms for about half of the available nights, and don’t know that customers like their neighborhoods equally and aren’t making choices on the basis of location.
“By doing nothing, Airbnb earns high commissions,” Birge says. “But if it reveals to the hosts the information that the customers are not choosing on the basis of location, the hosts might realize that by cutting prices, they’ll book more nights and increase their revenue.” Hosts in two equally desirable neighborhoods might compete with each other by lowering their per-night prices—and Airbnb’s commissions.
The second model the researchers analyze is “reveal and incentivize,” or RI, in which a platform gives sellers all the customer information it has up front and offers price-setting incentives such as the premium option extended by Airbnb. This model, the researchers say, provides the benefit of transparency and helps eliminate demand uncertainty that can lead to suboptimal pricing and revenue losses for platforms.
However, the researchers find, this model is also not ideal. If Airbnb gave its hosts all its information without limits, it could also work against the platform, Birge says. For example, “the hosts might end up competing too much against each other and drive down the commissions that Airbnb receives,” he says.
A third model—“strategic reveal and incentivize,” or SRI—melds the do-nothing and RI models into a more flexible policy. A platform could provide price-setting incentives and then decide on the most advantageous timing of information disclosure, or whether it even wants to disclose data at all.
Under this model, Birge says, Airbnb could first provide hosts with incentives, such as discounted commissions for prices that match the platform’s targets, while compiling information on consumer demand. The platform could collect data, at various points determining whether to reveal the information to vendors or continue the policy without disclosing anything further.
“The idea of this policy is that it captures the best of the other policies,” Birge says. “It looks most like do nothing when do nothing is best, and it follows RI when RI is best. So the main message is that it can adapt to different situations and end up applying the best policy for each.”
John R. Birge, Hongfan Chen, N. Bora Keskin, and Amy Ward, “To Interfere or Not to Interfere: Information Revelation and Price-Setting Incentives in a Multiagent Learning Environment,” Manufacturing & Service Operations Management, forthcoming.
Large retailers filling online orders from their brick-and-mortar stores must perform a balancing act.
The Equation: How Researchers Worked Out the Best Algorithm for Online Orders
User attempts to outwit algorithms can lead to less-optimal outcomes for everyone.
A Tool to Help Algorithms Cope with Human ‘Gaming’
The most influential rating system rests on some faulty calculations, affecting millions of people and billions of dollars.
Hospital Ratings Are Deeply Flawed. Can They Be Fixed?Your Privacy
We want to demonstrate our commitment to your privacy. Please review Chicago Booth's privacy notice, which provides information explaining how and why we collect particular information when you visit our website.