Airbnb’s Biggest Weapon Against Hotels: Machine Learning

Airbnb numbers among the Valley’s darlings, reportedly valued at $31 billion and actually profitable, the latter being a novelty among startups. It’s less beloved by the hotel industry. Although its share of revenue in the global hospitality market is only 3 percent, according to an analysis from January, it’s primed to increase that rapidly, with 18 percent adoption among travelers.

Airbnb doesn’t want to compete with the hotel industry merely on price (although clearly cost is a prime consumer concern). The company would prefer to also offer a booking experience that users will find more congenial and convenient. Toward that end, Airbnb continues to ramp up its investment in data science and machine learning.

Airbnb would obviously benefit from having the same advantage in travel that Amazon has in retail — you comparison-shop on its platform and nowhere else, instead of evaluating the options from multiple aggregators. Ash Fontana, a venture capitalist who specializes in artificial intelligence, pointed out that “if you’re a [travel] marketplace, you build up this data advantage” by knowing everything about the options that users are evaluating and how they’re reacting to them. By contrast, hotel chains don’t have direct or complete access to such a diversity of information.

In a way, aggregators like Priceline, Travelocity, and TripAdvisor should worry Airbnb more than the hotels themselves. Those entities often serve as the portals for users that make selecting a place to stay manageable — and cheaper than booking directly. Airbnb thinks that it has an additional advantage because its reviews are guaranteed to be legitimate, since they are tied to a booking and payment. Systems in which anyone can sign up and leave a review are theoretically vulnerable to abuse by competitors who want to sabotage one another or boost themselves.

One of Airbnb’s priorities is personalization at scale. “In Airbnb’s community, we have millions of guests and millions of hosts, and all of them are completely distinct,” VP of engineering Mike Curtis told Inc. “Each guest wants something different out of the travel experience that they’re going to have, and every host is offering a completely unique space.”

Therefore, search results aren’t generic, but rather based on your exact profile, your past behavior on the platform, and the behavior of users similar to you. “Search is the most central place that [matching guests and hosts] happens,” Curtis said. Not unlike Google’s search algorithms or Facebook’s News Feed, Airbnb weighs many factors when deciding which listings to show you. And actions speak louder than words.

Curtis explained, “We can look at things like, how long are you planning to stay? How soon are you planning to stay?” But those are just the basics. Airbnb can also glean your preferences from your behavior: “What common amenities do [the listings you look at] have? Are they in an outdoorsy or a more urban setting? What neighborhoods are they in? We have hundreds of signals that go into the search rank.”

While you’re searching, Curtis said, the platform is “re-ranking results and presenting the listings that are best for you.” Data about how you react, including post-trip ratings and reviews, all goes back into the system. (Curtis noted that sentiment analysis of review text is something Airbnb has thought about but not yet implemented.)

Consumers navigating a platform like Airbnb experience a catch-22, said Fontana: “Marketplaces are most useful when they have a lot of volume, because you can find exactly what you want, but marketplaces are also the most time-consuming and annoying when they have the most volume.”

Machine learning is the way to reconcile the problem, Fontana said, since “recommendation systems and personalization are core use-cases” of the technology.

Airbnb’s investment is paying off. One of the primary success metrics is the platform’s conversion rate — how many people make a booking. Airbnb also examines how long it takes a user to choose where he or she wants to stay, optimizing for quicker decisions. Curtis said that Airbnb’s conversion rate “has gone up pretty sharply” in the past few years. He declined to specify a percentage, but maintained that “it’s a significant improvement.”

Curtis did reveal that introducing a deep neural net to the search-ranking system boosted Airbnb’s recent conversion rate by 1 percent. “One percent may not sound like a lot, but as you can imagine, a 1 percent increase in conversion rate compounds over time,” he explained.

Curtis doesn’t just see Airbnb’s use of cutting-edge technology in terms of ROI, but also as a hopeful sign regarding innovation’s impact on the future. He pointed out that Airbnb provides a nontrivial chunk of income to many of its hosts.

“In a world where there is high likelihood of more jobs being automated away by these incredible technologies that are coming out, there needs to be companies that are thinking about, ‘How are we going to leverage this technology in a way that enables people to work?'” he said.


Article by channel:

Read more articles tagged: Machine Learning