News & Events
With more than 1.5 million listings worldwide, featuring everything from an Irish preparatory school that sleeps 70 to a tiny cabin in upstate New York that barely has room to lie down, Airbnb enables travelers to find accommodations that suit their tastes to a T.
But how do customers know they are getting the best possible price?
Now, thanks to a data-driven tool developed by two students in the computational science and engineering master’s program offered by the Institute for Applied Computational Science (IACS) at the Harvard John A. Paulson School of Engineering and Applied Sciences, Airbnb customers can have a better handle on how—and when—to haggle with hosts.
Jack Qian, S.M. ’16, and Qing Zhao, S.M. ’16, collaborated with Michele Invernizzi and Giovanni Conserva, students at the Italian technical university Politecnico di Milano, on the project for their capstone class taught by Pavlos Protopapas, Scientific Director for IACS. The project provided a cross-cultural experience for the SEAS students, who traveled to Italy to work with their teammates, hosted the Italian students when they visited Harvard, and continued to collaborate remotely throughout the term.
“Most people are not even aware that there is room to negotiate prices with Airbnb hosts,” said Qian. “The goal of our project is to give the consumer more power to get a better price.”
To tackle that complicated issue, Qian and Zhao took a two-pronged approach. They developed a machine-learning algorithm that can predict whether an Airbnb host would be willing to negotiate, and also recommend a specific asking price that host would be most likely to accept.
The students analyzed a massive amount of Airbnb data, using statistical inference to determine which pieces of information would be most important for negotiating price. For example, booking dates are weighed heavily when determining the likelihood of successful negotiation. If a guest’s booking fits perfectly into a period between two bookings, the guest will have more power to negotiate a lower price, since the host likely wants to keep the property booked for as many consecutive days as possible.
Qian and Zhao built a number of other factors into the algorithm, such as the location of the property, the popularity of the host, and whether the booking was made weeks in advance or only a few days before. After their model output the relative importance of each factor, they studied those outputs to ensure the mathematical results made intuitive sense.
“We didn’t want to develop a black box where we just toss information in and then a result comes out,” Qian said. “The process is a combination of art and science. We need to decide what factors are important, but those decisions must be supported by the metrics outputted from our model.”
They incorporated the polished algorithm into a user-friendly website which lets users search Airbnb listings for a specified date and location. The site identifies hosts who are likely to be open to negotiation, and also suggests an asking price that maximizes the chances the host will accept. Guests connect to the Airbnb site to contact hosts and complete the booking.
The algorithm improves by more than 150 percent the level of precision with which hosts who are willing to negotiate are identified. Only about 20 percent of Airbnb hosts identified through a standard search are willing to negotiate, but that increases to more than 55 percent once their model narrows down the host list. That improvement in precision is impressive, Zhao said, since predicting the willingness of hosts to negotiate involves a number of factors that can vary widely on a case-to-case basis.
“From a business point of view, we want our tool to work for customers, but also not annoy hosts,” she said. “We want to reach a win-win situation, so we have to carefully think about how our tool would benefit both Airbnb and the hosts. We have to reach a balance to make everyone happy.”
While they built their model with the customer in mind, it could benefit hosts, too, since they can clearly see how reducing the price of their listings in certain situations could improve the occupancy rate. Now that they have developed a proof of concept, their next step involves further refining the model to improve precision, said Qian.
“It has been nice to see that the lessons we are learning in class can be applied to real life problems,” he said. “Rather than thinking of this as a class project, we have been focused on the customer and adding value to the community, and that has been very rewarding.”
Topics: Computer Science
Cutting-edge science delivered direct to your inbox.
Join the Harvard SEAS mailing list.