Here’s the thing about AI: It’s pretty much the only tech breakthrough in the past decade, maybe even longer, that demonstrates touchable, tasteable, real-life, concrete, measurable ROI.
And the measurable impact that machine learning has had on Airbnb’s unique technological challenge — creating great matches between guests and hosts — has been “profound,” says the company’s VP of engineering Mike Curtis, who’s a featured speaker at MB 2017 coming up July 11-12.
Airbnb connects millions of guests searching for the right place to stay and millions of hosts offering distinct spaces. Airbnb’s unique technological challenge is to personalize each match between guest and host.
The goal is to create a “great match between a guest and a host that’s going to lead to a great experience out there in the real world,” says Curtis.
Helping guests find the perfect place
A big part of the magic lies in personalizing rank search results for guests coming to the site.
Initially, search rankings were determined by a set of hard-coded rules based on very basic signals, such as the number of bedrooms and price. And because they were hard coded, the rules were applied to every guest uniformly, rather than taking into account the unique values that could create the kind of a personalized experience that keeps guests coming back.
Airbnb learned over time that machine learning could be used to offer this personalization, Curtis said. Airbnb introduced its machine learned search ranking model toward the end of 2014 and has been continuously developing it since. Today Airbnb personalizes all search results.
Airbnb factors in signals about the guests themselves, as well as guests similar to them, when offering up results.
For example, guests provide explicit signals in their search — the length of stay, the number of bedrooms they need. But as they examine their search results, they may show interest in similar, desirable attributes that the guests themselves might not even notice.
“There’s a bunch of other signals that you’re giving us based on just which listings you click on,” Curtis says. “For example, what kind of setting is it in? What kind of decor is in the house? These are things Airbnb can use to feed into the model to come up with a better prediction of which listings to show you first.”
The company pulls well over a hundred signals into the search rank model, Curtis says, and then the machine learning algorithm figures out how all the signals interact, to produce personalized search rankings.
“The beauty of machine learning is that it comes up with a set of weights that we should put against each of those signals that it believes comes up with a good prediction of what’s likely to end up being a good outcome or a good match,” he adds. “Or in other words, you book that place.”
Helping hosts nail the perfect price
Airbnb has also invested in a predictive pricing model for hosts.
“One of the challenges that our hosts face is that it’s pretty hard for them to know how to price their listings,” he says. “So you find hosts doing all these things like looking up how much hotel rooms cost in the area or looking at listings around them to try to get a sense of how much they should charge for their listing.”
But it’s difficult for a host to keep up with the balance of supply and demand, and the myriad forces that go into market pricing.
For instance, their space could be worth more because it’s in bigger demand when a big conference is in town, or worth less because it’s Memorial Day weekend and a lot of people are out of town and supply is more plentiful.
“So what we set out to build was something that could come up with a prediction based on all of the historical data we have on travel patterns that can tell us the probability of any listing being booked on any given night at any given price up to 12 months into the future,” Curtis explains.
Based on what their occupancy interests are, hosts can choose to take those suggestions by default so that Airbnb manages the pricing of that listing. On a daily basis, the algorithm recalculates price predictions for all listings around the world, up to 12 months in the future.
“We’re always iterating on this stuff,” Curtis says. “We’re always experimenting and looking to see, is this helping hosts successfully get more business? And is it helping guests successfully find the right matches for the right place for them faster and easier? Because that’s what we want to do.”
Those two data products — machine learned search ranking and price prediction algorithms — are two of the most impactful incremental business growth drivers, Curtis says.
The other thing about AI: It’s that special kind of tech breakthrough that doesn’t just affect bottom lines — it’s also affecting the real world.
“The problem that we’re solving here is so fascinating,” Curtis says. “Because when you think about it, all of the technology that we’ve built, all the online tools, everything else — none of that is actually the real product. The real product is out in the world, human interaction in the physical world, and everything that we do in technology and online is to facilitate more great offline experiences and connections between people.”
“It’s amazingly cool,” he adds, “and it gets better all the time.”