How Airbnb Solved The Mystery Of Predictive Pricing

You’re not a real estate agent, so how can you know how much that spare bedroom is worth? Airbnb did some algorithmic sleuthing to help you know.

How Airbnb Solved The Mystery Of Predictive Pricing
[Image: Flickr user Jeff Kubina]

When product lead Dan Hill watched new Airbnb users make their first furtive listings, he noticed a pattern. They’d glide through the standard questions of address and room size, but then get stuck on the exact same item: price.


“It seems like an innocuous little text field,” Hill says.

Stumped for what they should make their nightly rate, users would frantically open up other tabs, look up neighboring hotels, or just punch in a number and pray. There was way more friction there than anywhere else in the first-time listing process. Clearly, Airbnb had a user experience problem.

“We saw this transition,” Hill says. “People are used to the routine form-filling online, entering credit card details and such, and suddenly they hit (the room rate) and say, ‘oh, I’ve never thought of that question.'”

This was back in June 2012. Hill had just come to the company; he was the cofounder of Crashpadder, a similar space-sharing service that Airbnb acquired. This was three years after Airbnb came out of the seed accelerator Y Combinator; two years before they’d be on the cusp of becoming the world’s largest hotel chain.

The user’s not knowing what they should charge signals the change that Airbnb represents. As Hill says, since the beginning of the web, we’ve been able to buy things through ecommerce, then the social wave hit and we could poke each other online, but then companies like Airbnb came on the scene, where real people are interacting offline.

“When you get into the offline space,” Hill says, “you get into all these interesting challenges that people have never considered, like how much should you charge, should you provide towels, what happens if you don’t speak the same language as someone staying with you?”


These questions require technological solutions. For the room rate, Airbnb set out to create predictive pricing. Rather than force users to summon their inner real estate agent, Airbnb set to crafting an algorithm.

The first element is the listing itself: is it a private room, shared room, or whole apartment? Does it sleep 10 people or one? Then there’s seasonal demand: Is the Olympics or Super Bowl in town? Is it summertime at a ski resort? Then there’s the location: Is it in the center of the city? Is it in a hip neighborhood?

But the thing about places is that they’re complex. In London or Berlin or New York, you’ve got rivers and highways making everything messy. Plus a flat on one side of the Thames could be way more desirable than another one 200 meters away on the adjacent shore. To understand locations, Airbnb would have to get more precise.

To that end, they acquired a company called NabeWise that was cataloging and codifying the world’s urban neighborhoods. That turned into Airbnb’s Neighborhoods feature. In there you’ll find gorgeous photography and breathless descriptions of these places. But behind that beauty is a lot of labored breathing.

“Behind the scenes they do a very accurate polygons, boundaries around every single one of these neighborhoods,” Hill says. “All of a sudden we have access to first 10, then 50, then god knows how many cities around the world we’ve drawn exact neighborhood boundaries–around Kensington or Chelsea or around the Mission or the Marina. That allowed us to really accurately describe where a listing is.”

Then, deeper into the back end, Airbnb invested in their data infrastructure, warehousing, and storage. Coupled with upgraded technology, the company could now do heavy-duty calculations that enable immediate results for users. Which lead to results for Airbnb.


When Airbnb first did an experiment with the predictive pricing, users who chose to use the suggestion price got three times the number of bookings than the control group–making a site-wide rollout an easy call. The solution to the what should I charge?! problem, then, was to get nuanced with neighborly boundaries and build the processing muscle to allow for a smooth user experience.

“The analogy we often use is a swan on a lake, when you look at a swan, it’s gliding perfectly across the water’s surface, it’s so beautiful and people write poems about them, but underneath the surface, his legs are frantically kicking all over the place,” Hill says.

“That’s what we strive for: digesting your problems to the simplest kernel of an idea and solve for that idea–and the user, they should never even have to consider, that shouldn’t even come up as a problem. They should just glide through that experience.”


About the author

Drake Baer was a contributing writer at Fast Company, where he covered work culture. He's the co-author of Everything Connects, a book about how intrapersonal, interpersonal, and organizational psychology shape innovation.