Surabhi Gupta is a Director of Engineering at Airbnb. She leads Engineering for the Homes business which includes Growth, Search, Host and Business Travel. Prior to Airbnb, she was a software engineer at Google where she worked on web search ranking and the Google Now team on predictive search.
Airbnb’s astronomical early adoption rates may have been based on the hosting engine’s competitive pricing and ability to undercut the hotel industry. However, as the platform expands into tours, activities, restaurant bookings, and concierge services, Airbnb’s future growth will rely on their ability to offer guests ultra-personalized booking services. This is in line with similar machine learning trends among other firms.
At least, that’s the idea behind the firm’s expansive deployment of machine learning, which Surabhi explains has been in use since the platform’s early days.
“AI is not new at Airbnb,” she says. “We have used machine learning techniques for years—for everything from search ranking to fraud detection to sentiment analysis—but our path to unlocking those abilities was a methodical one.”
Airbnb believes its continued investment in AI and machine learning capabilities will take their core and expanded services to the next level. Indeed, they already use Automated Machine Learning to accelerate data science projects across the platform, from benchmarking challenger models and diagnostics to exploratory data analysis. Now, machine learning will generate new user experiences for Airbnb guests.
“Machine learning—which falls under the umbrella of AI—is incredibly critical to the future of business. We believe Airbnb’s approach is unique when compared to what other tech companies in this space are doing. We are not looking to replace / automate people, but rather focused on developing technology that enables and fosters real-life human connections and experiences.”
Machine Learning Is Transforming Airbnb’s Search Ranking
With over 200,000,000 guests in total, and over 3 million listings, Airbnb has plenty of data with which to leverage this technology—and it’s driving significant growth. Surabhi explains that AI is vital to unpicking this data. “We have an immense amount of data at Airbnb, and have put AI and machine learning systems in place to help us understand what is important—and what is noise—to help us make better decisions for our users and our business,” Surabhi explains.
“We match guests not just with hosts and listings that align with their preferences, but also with neighbourhoods and experiences that meet their needs and interests, which makes for a better consumer experience through and through.”
“With the vast amounts of data being generated every day and the computing power to process it, machine learning is the most significant tech breakthrough in the past decade that demonstrates real-life, concrete, measurable ROI. It allows us to generate insights from the interactions users have with our product and we are able to use these insights to help build a better experience for the end user.”
One of the most critical applications for machine learning is also Airbnb’s most central feature: search ranking. “Every guest and every host on Airbnb is completely unique, with their own set of preferences,” says Surabhi. “The search results we display are personalized to the listings and experiences we predict will be best for that guest, as well as the trip characteristics that are right for the host.”
Ultra-Personalized Travel Options
Airbnb achieve this level of personalization by inferring guest preferences based on their interactions with the app. As guests plan their trips by engaging with listings and making inquiries, Airbnb quietly notes their preferences in terms of factors such as location, price, amenities, and more. “It also turns out that at the time a guest is searching, we can predict with reasonable accuracy if they would give a particular home a good star rating or a poor star rating if they were to stay in that home. We use this in our search ranking model, going beyond what just generates the most bookings, but what generates the highest likelihood of a great experience offline.”
Lately, the platform has begun experimenting with personalization that incorporates much softer attributes like the look and feel of a home. “Our goal is for guests to find the trip that is right for them as quickly as possible. Users will see more personalized results based for their upcoming travels. As we’ve refined our search rankings, we’re enabling guests to have a more fulfilling experience—and facilitating connections with our hosts, who also benefit from a more efficient booking experience.”
“Matching has personalization at its heart, and the new system is designed to understand travelers’ preferences,” Surabhi explains. “We match guests not just with hosts and listings that align with their preferences, but also with neighbourhoods and experiences that meet their needs and interests, which makes for a better consumer experience through and through.
Improved matching = new opportunities?
Personalization obviously has its benefits for individual customers, but its effects go much further. The expanded use of machine learning accompanies a wider shift in Airbnb’s business model, and with it, the potential for a concurrent renewal of stakeholder relations. The firm’s disruptive model has faced a number of regulatory challenges in previous years from local governments and residents that have struggled to quickly adapt to the rapid market shifts that have been unleashed. By expanding beyond accommodation towards offering travelers the ability to organize and personalize every aspect of their trip using the platform, improved matching could lead to a more even distribution of opportunities for local stakeholders and businesses. Surabhi explains that this is because it “opens up new neighbourhoods and local businesses for tourists and visitors that may be off the beaten path, or in lesser-known areas of the city.”
Using machine learning, Airbnb also aim to provide a better deal to hosts by matching guests with them according to their mutual preferences and past experiences. “Targeted matches take both host and guest preferences into account—our search tool does not just reveal listings that align with guests’ past behaviour and preferences, but also listings that, if they request to book, a host would be more likely to accept based on the trip characteristics. This results in happier guests and hosts, as well as more matches; ensuring the healthiest distribution we at Airbnb can see in host and guest matching.”
Airbnb at next week’s AI Summit San Francisco
Regarding next week’s AI Summit, Surabhi had this to say: “I’m looking forward to meeting other entrepreneurs and executives that are working hard in the AI and machine learning space. As these technologies continue to grow, it is critical that we learn and work together so that when we build the best technology, it has the ability to improve lives, build communities, and further human connections.”
Surabhi Gupta is a Director of Engineering at Airbnb. She will be delivering a keynote speech at the AI Summit San Francisco 2017 entitled ‘Using machine learning to improve matching in marketplaces’.