Search is so engrained in our daily internet lives that we sometimes forget to stop and think about how it works. Often, when digging for a website on Google or finding a video on YouTube, search is straightforward: type a few keywords into the box, and sift through the results to find what you’re looking for. But what if you’re searching for something so specific that just a couple of keywords and extra parameters can’t filter it? Or, if you don’t even know what you’re searching for?
Those are the challenges that the search team at Airbnb is thinking about. In the past few months, the room and apartment sharing company has worked on tuning its product to support different ways people book vacations. First, the company unveiled its last-minute booking product in San Francisco and Los Angeles. Now, the company is also looking into a Weekend Getaways feature, designed to inspire and entice users who would otherwise not be compelled to book a trip to make a last-minute excursion nearby.
“In Google, there is such a thing as the objectively best result,” Maxim Charkov, head of search at Airbnb, told me. “For Airbnb, people have very different preferences. In many cases, we don’t know know what those preferences are, so we have to iterate and find them.”
In order to make this happen — while managing Airbnb’s shifting host inventory and also pushing users to book rooms and apartments that suit their needs — Airbnb’s search team is trying to rebuild search from its foundation.
How it works
The main driver of search at Airbnb, Charkov said, is simply location: users type in where they want to go and how many people (if any) will join them. The company ensures that users not only get accurate location results, but also receive suggestions for locations they did not search for — one example, Charkov said, is when Airbnb serves a popular listing in Aptos for users searching in Santa Cruz. But the results that come from that search are much more complex, as Airbnb factors in dozens of factors to measure the “attractiveness” of each location, including photos, quality, ratings and cost of the listing.
“Quality can be drastically different. Airbnb is a user-generated marketplace, and there are a lot of gems, and we have to find them,” Charkov explained. “We think of quality as simply a measure of how attractive the listing is, independently of who you are. It’s just how good it is.”
From there, Airbnb takes a deep look at behavior to further deduce the quality of listings. This behavior starts with actions from users looking to book — especially testing when and how users click on listings — but become much more interesting on the host side of the spectrum. Airbnb closely studies what it calls a “conversion rate” — the likelihood of an inquiry to turn into a full booking. The conversion rate is dependent on a multitude of factors and some, like when a host has to suddenly black out days for an emergency, can’t be measured by the company. However, it does look at certain key pieces of information that compel a host to book a guest — such as the number of guests involved, the length of the stay, the turnaround time between other bookings, and even the specific season the booking is for — to surface better listings on the user end. This becomes particularly important as the company expands its last-minute booking feature beyond Los Angeles and San Francisco.
Charkov says that Airbnb can algorithmically surface listings from user behavior and other quality markers, but increased search power comes from participation from the users — both hosts and guests. Currently, Airbnb has 650,000 listings in 34,000 cities across 190 countries, and more than 50 percent of the company’s business is in Europe. As Airbnb’s popularity grows all over the world — it said in a blog post in March that it is experiencing triple-digit growth in all European countries — there is potential for a lot more diversity in available rooms and willing guests. To help unearth those preferences, Charkov and his team are running tests to make Airbnb less like search and more of a matchmaker.
“One of the biggest problems the team has right now is understanding signals beyond quality,” Charkov says. “What we are trying to do now is learn how to tell apart these listings. What does a listing represent, and is it good for a certain group of users?”
One of the ways that Airbnb’s search team tested enhanced preferences took advantage of the company’s acquisition of neighborhood information platform Nabewise to help better characterize neighborhoods in given cities. From there, the company turned to the data it gathered from user reviews on individual listings to pick out certain keywords like “food,” “nightlife” or “outdoors” to parse which listings were closer to ideal features that travelers normally look for. The results ended up being much higher quality than information offered by hosts themselves, and could lead to better search results in the long run.
“Anyone can write absolutely anything they want in the description, but we want to get feedback from people who actually stayed there,” Charkov added.
It’s those extra signals that separate particular listings from others in a similar location, and could help users find the perfect place to stay in the long run. Charkov says that as Airbnb scales, the goal of the search team is to break up listing results into reasonable chunks. Using as much information as possible, the company can then better match users with a smaller set of listings that suit their needs without overwhelming them.
“I see that matching and personalization will be more critical than it is for Airbnb now,” Charkov said. “We will need to be much better at not just matching the user behind the scenes, but also explaining why we are matching the user with a given set of listings.”