This weekend, tens of millions of Americans are heading off to the beach, the lake, the mountains, or wherever the barbecues and beers await. And for many holiday travelers these days, that means scouring Airbnb to find that perfect oceanfront cottage that sleeps eight, comes with a washer, dryer, Wi-Fi, and free parking on the premises.
But what most people won't realize when they nestle into their respective crash pads this July 4th is just how complex that search process really is.
Airbnb, of course, is not one of the giants of search. Google and Amazon have it beat by almost any measure. But unlike either of those companies—or Facebook, Instagram, and Twitter, for that matter, which have all emphasized search in recent years—Airbnb faces an altogether unique set of challenges, most notably, the fact that its search results don't simply reflect websites or photos or products. They reflect people—people who may be renovating their homes, people who don't want to accommodate a two-day rental in the middle of the summer, people who don't check their email, or people who might want to spend the holiday weekend at their own darn lake house, thank you very much.
And yet, it's still Airbnb's job to predict the whims of these hosts to ensure that guests can find a place to stay every time. That means Airbnb can't simply surface all its listings in a given area, no more than Google could arbitrarily serve up every web page in random order. Both businesses depend on users finding the right answer fast.
"You always need to match supply and demand, and in our case, the supply is completely unique. You're talking about hosts and their homes," Airbnb vice president of engineering Mike Curtis said on a recent visit to WIRED's New York office.1 "So the matching problem between what’s the right host for the right guest is a pretty complex one."
To solve that problem, Airbnb is increasingly using machine learning to understand the habits and preferences of both its guests and its hosts in order to make the most relevant matches. Each time a user searches for accommodations on Airbnb, the company runs that search through a model to determine which hosts are most likely to accept. The model factors in variables like the duration of the stay and the gap between this latest potential booking and the host's last booking, among other things.
In a test of the model, Airbnb's researchers found that ranking listings based on the host's likelihood to accept led to a 4 percent lift in actual bookings. So Airbnb has been using the model ever since.
Meanwhile, Curtis says the company is also working on ways to explicitly collect preferences from hosts that can be layered on top of the machine learning model. For instance, knowing whether or not hosts can accommodate last minute bookings would be critical to capturing users who have limited time to find a place to stay and might otherwise turn to rival services like HotelTonight. "As we move more to instant booking, it gets even more important that we understand those host preferences," Curtis says.
The last piece of the puzzle is using technology to understand user preferences beyond the filters they've explicitly selected. From click patterns, the system could learn, for instance, that a given user always favors brightly lit spaces. Or perhaps they're accessing Airbnb through its integration with Concur, a travel and expense management system for businesses. That would indicate to the system that the user is a business traveler who would likely require basic amenities like on-site laundry and Wifi.
Curtis admits that Airbnb's prediction models around guest behavior are still pretty limited, but he says that they will be a continued area of focus for the company over the coming year. And they should be. For tech companies these days, the pressure to deliver instant satisfaction to users has never been greater. It's why Google is suddenly cherry picking search results to expand at the top of the page and why Amazon labors over its recommendation engine to surface products you didn't even realize you needed. Sophisticated search helps these companies clinch customers quicker than competitors can steal them away, making it a critical skill for Airbnb, and any other online company, to master.
"You can't have somebody search for a place to stay for the weekend in New York and have to go through tens of thousands of places to find the one they want," Curtis says. "We need to get smarter and smarter."
1. UPDATE 9:21 am ET 07/06/2015: An earlier version of this story misstated Mike Curtis's title. It has since been updated.
