BETA
This is a BETA experience. You may opt-out by clicking here

More From Forbes

Edit Story

How Airbnb Uses Big Data And Machine Learning To Guide Hosts To The Perfect Price

This article is more than 8 years old.

Airbnb wants its hosts to set their own prices. But the home-sharing company, armed with billions of data points, is nevertheless starting to nudge hosts toward prices that earn them -- and Airbnb -- more money.

Price Tips, a feature that the company announced Thursday at its OpenAir conference in San Francisco, is a constantly updating guide that tells hosts, for each day of the year, how likely it is for them to get a booking at the price they've currently chosen. Hosts can glance at a calendar and see what dates are likely to be booked at their current price (green) and which aren't (red), and they can get price suggestions as well. When hosts price themselves within 5% of the suggested price, they are "nearly four times" as likely to get a booking as when they don't, Airbnb said.

The price tips are presented in a easy interface laid over a complex process -- one that crunches everything from the day of the week to the specific neighborhood of a listing and surfaces patterns between latitude, longitude and key words like "beach," two Airbnb employees explained Thursday afternoon at the conference.

Before the price tips, hosts received general price suggestions when they added a listing. And many are savvy enough to know when major events will lead to a spike in demand. But hosts (and Airbnb, which takes a cut for service fees) might lose money on highs and lows that they aren't aware of, like underpricing themselves during a huge conference or asking too much -- and getting no bookings at all -- during the low tourist season.

Getting the most out of their listings will likely become more and more important to Airbnb hosts, especially in San Francisco, where city officials are debating a cap on the number of bookings a listing can have per year.

Such specific price recommendations -- within 5% of Airbnb's algorithmic result -- also blur the line between Airbnb as a marketplace and as a more controlling actor. Unlike some on-demand marketplace companies like Uber, Lyft and Homejoy, Airbnb has until now stayed away from setting prices for its hosts. (A potentially smart move, since marketplaces that do control pricing are getting in legal hot water.) But its price tip feature suggests that Airbnb can't resist getting a little closer.

Price Tips in action. (Courtesy Airbnb)

Airbnb's price suggestion engine, which took months to develop and pulls on five billion training data points, has two main components: modeling and machine learning, explained Airbnb data scientist Bar Ifrach at Thursday's conference. The model pulls together what Airbnb's huge data set can reveal about a listing's best price based on things like its neighborhood and the size of the listing.

Some trends are intuitive. Any time a city has a big event -- in the example, Ifrach pointed to South By Southwest and Austin City Limits -- best prices will jump. Sometimes events raise prices citywide, but other times it varies greatly by neighborhood. In a demand map of San Francisco, neighborhoods like SoMa and the Mission are more likely to be booked than the outer corners of the city (and the Tenderloin, which was the least-likely spot on the whole map). Even two spots just a few blocks away from each other might have significantly different demand.

The value of amenities also varies greatly. Wi-fi makes a listing in San Francisco more likely to be snapped up, whereas hotter climates will see many fewer bookings if they don't offer air conditioning -- though that also depends on the season. Sundays are the least likely day of the week to get a booking, and the rate rises slowly throughout the week until Saturday, the most popular day.

More surprisingly, the number of reviews can play a huge role in the likelihood of a booking. A listing with no reviews is seen as uncertain, but just one good review makes it much more likely to be booked. Beyond that, each review helps a little more, but doesn't add nearly as much. Having one three-star (which, despite being the middle of the pack star-wise, is seen as a not-great review at Airbnb, Ifrach said) can hurt a listing.

But those are the factors that Airbnb knew to model for. To find even more relationships between listings and the prices they can command, Airbnb developed Aerosolve, a machine learning package that it released open source on Thursday.

With Aerosolve, Airbnb can surface new patterns that it then uses to better understand what makes a listing command a certain price, explained Hector Yee, an Airbnb engineer. For example, the model highlighted that listings at a certain latitude and longitude were commanding good prices and were often using the word "sabbia." Turns out it was Playa del Carmen, a popular beach town in Mexico, and "sabbia" is "sand" in Italian -- another piece of information that can Airbnb can relay to local hosts in the form of a price tip.

Airbnb hosts "don't have an army of analysts," Ifrach said. "We're trying to equip, empower our hosts so they are able to price the listings, get bookings, and do it seamlessly, effectively, so we have more hosts and more stays on Airbnb."

 

Follow me on TwitterSend me a secure tip