Airbnb: Utilizing Machine Learning to Optimize Travel
Airbnb utilizes machine learning to personalize search rankings for guests and to optimize pricing for hosts.
“Every time you interact with an Airbnb app or the website, you’re interacting with machine learning in some way or another.”
– Mike Curtis, VP of Engineering, Airbnb [1]
Founded in 2008, Airbnb has quickly grown into one of the largest players in the travel industry. Airbnb’s business model is simple: it is a global, online marketplace that connects travelers who are looking for a place to stay with hosts who are looking to rent unique accommodations. The company has recorded more than 400 million total guest arrivals, 5 million listings, and 2 million guests per night on average [2]. Airbnb has fueled this growth by utilizing machine learning to solve a complex problem: matching guests and hosts. Despite limited available data for both parties, Airbnb has successfully integrated machine learning into many aspects of its product development process.
In the near-term, Airbnb is focused on utilizing machine learning to (1) personalize search rankings for guests, and (2) optimize pricing for hosts.
Personalized Search Rankings
During the early days of Airbnb, search rankings were determined by a handful of hard-coded, basic variables such as dates, duration of stay, and price [1]. However, as Airbnb scaled its number and tenure of users, it collected valuable data that could be used to predict listing preferences [3], [4]. During an interview with VentureBeat, Mike Curtis, VP of Engineering at Airbnb, noted, “There’s a bunch of other signals that you’re giving us based on just which listings you click on. 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.” [1]. While Airbnb launched the personalized search ranking model in 2014, the product has and will continue to evolve over time. Particularly, as the company continues to launch new product offerings (i.e. Experiences), it will capture new data, refine its algorithms, and become even more accurate at predicting user preferences.
Price Optimization
One challenge that hosts have consistently communicated to Airbnb is how challenging and time consuming it is to determine nightly rates [5], [6]. To help address this issue, Airbnb developed a proprietary model to predict maximum revenue per night for listings. This model utilized machine learning to predict the probability of bookings at various price points. The model is based on both external factors (such as hotel rates, seasonality, market popularity, or local events) as well as control inputs from hosts (minimum/maximum prices, frequency of hosting, etc.) [7]. Airbnb combines this data to predict appropriate pricing for a listing. This feature is called “smart pricing” and today uses more than 70 factors to determine optimal nightly pricing [8], [9].
Exhibit 1: Example “smart pricing” [10]
Looking Ahead
While personalized search rankings and price optimization are two near term initiatives, there are many other ways that Airbnb can utilize machine learning in the medium term. The VP of Engineering at Airbnb has identified several initiatives, including: (1) using images to improve search rank, and (2) improving reviews by using natural language processing [11]. Airbnb can use image classification to improve search rankings by ordering photos based on what guests care about the most (i.e. bedroom).
Exhibit 2: Example of photo classification [12]
Airbnb can also use natural language processing to improve guest reviews. For instance, reviews often focus on the city that the guest visited rather than the quality of the accommodations. Through using natural language processing, Airbnb can rank reviews based on quality, content and relevance.
Recommendations
While Airbnb’s management has developed key focus areas for machine learning over the next few years, there are many other opportunities for the company to use machine learning. In the near term, Airbnb could further advance its search rank algorithm by using machine learning to analyze guest reviews. Through analyzing reviews, Airbnb could capture valuable data about positive and negative guest experiences. This data could be used to inform the search rank: if another guest had a similar review of a listing, Airbnb can promote or demote that listing based on the guest’s former reviews. In the long term, Airbnb’s vision is to own the entire travel ecosystem: lodging, experiences, transportation and services. To achieve this “one-stop-shop” model, Airbnb needs to “own” the customer long before he or she begins booking the trip. Airbnb should use machine learning to understand the trip ideation process as well as what other services users demand.
As Airbnb amasses more and more users, this data-driven approach will become increasingly complex and increasingly powerful. Can machine learning alone propel Airbnb through its next phase of growth?
(777 words)
Sources:
[1] VB Staff, “Airbnb VP talks about AI’s profound impact on results,” Venture Beat, June 14, 2017, https://venturebeat.com/2017/06/14/airbnb-vp-talks-about-ais-profound-impact-on-profits/, accessed November 2018.
[2] Airbnb, “Fast Facts,” https://press.airbnb.com/fast-facts/, accessed November 2018.
[3] Amelia Heathman, “How AI is powering Airbnb’s mission to change how we travel forever,” London Evening Standard, April 17, 2018, https://www.standard.co.uk/tech/airbnb-artificial-intelligence-21st-century-travel-a3816336.html, accessed November 2018.
[4] Malay Haldar et al., “Applying Deep Learning To Airbnb Search,” Aibnb, Inc., https://arxiv.org/pdf/1810.09591.pdf, accessed November 2018.
[5] Stephanie Pandolph, “Machine learning is driving growth at Airbnb,” Business Insider, June 16, 2017, https://www.businessinsider.com/machine-learning-is-driving-growth-at-airbnb-2017-6, accessed November 2018.
[6] Airbnb, “How we used host feedback to build personalized pricing tools,” https://airbnb.design/smart-pricing-how-we-used-host-feedback-to-build-personalized-tools/, accessed November 2018.
[7] Airbnb, “Machine Learning in Matching & Marketplaces | Tech Talk | Airbnb,” YouTube, published August 29, 2017, https://www.youtube.com/watch?v=XiZJfBQqvdI, accessed November 2018.
[8] Harriet Taylor, “Airbnb launches ‘Smart Pricing’ for hosts,” CNBC, November 12, 2015, https://www.cnbc.com/2015/11/12/airbnb-launches-smart-pricing-for-hosts.html, accessed November 2018.
[9] Airbnb, “What’s Smart About Smart Pricing,” https://blog.atairbnb.com/smart-pricing/, accessed November 2018.
[10] Sharan Srinivasan, “Learning Market Dynamics for Optimal Pricing,” Medium, https://medium.com/airbnb-engineering/learning-market-dynamics-for-optimal-pricing-97cffbcc53e3, accessed November 2018.
[11] Airbnb, “Sharing More About the Technology That Powers Airbnb,” https://press.airbnb.com/sharing-more-about-the-technology-that-powers-airbnb/, accessed November 2018.
[12] Shijing Yao, “Categorizing Listing Photos at Airbnb,” Medium, https://medium.com/airbnb-engineering/categorizing-listing-photos-at-airbnb-f9483f3ab7e3, accessed November 2018.
This is an excellent essay and throws light on some of the key benefits of incorporating ML in modern-day applications (AirBnB did not even exist decades ago when ML was established as a theory). I firmly believe that AirBnB can unlock more potential through the use of ML in ‘lodging’. However, to be able to make headway into the associated areas (to become the one-stop-shop), AirBnB will first need to establish itself in those areas to be able to gather data effectively. A direct acquisition would prove helpful in that regard.
For the vanilla lodging solution, it is important to note that user choices are influenced by a negativity bias in the reviews – reviews that are negative are trusted more than the positive ones. If the website actively tries to manipulate the reviews, then the trust between the user and the platform is compromised, affecting the overall credibility of the platform. AirBnB should be cognizant of these risks before opting to use ML to influence user choices.
I do believe that the likes of Airbnb and VRBO are the future of hospitality and tourism. I think it is interesting what they are doing by using machine learning to put pictures of the rental in a different order that may attract more renters. However, I do have a concern with the AI matching guests and hosts. With machine learning if there is bias in the input data, there will be bias in the output data. As evidenced, by Airbnb’s partnership with the NAACP back in July 2017, they know they have inherent issues with bias that they are working to resolve
(https://www.theverge.com/2017/7/26/16037492/airbnb-naacp-partnership-racism-diversity-hosts).
So how does Airbnb combat this issue? Can they start blinding the names and genders of the guests from the algorithm, so they aren’t judged based on race or sex? Can they teach the algorithm to root out inherent bias? It will be interesting to see how Airbnb manages these political controversies moving forward.
Great essay! I think you’ve made an excellent observation about Airbnb’s mission to own the entire travel experience. As they continue to build out their offerings, I believe it would be beneficial for them to partner with airline companies or already established travel websites that alert consumers about deals. The normal progression of someone looking at their website is because they just looked at a travel method right before or intend to look at the travel method soon after. A partnership with a well established company can 1) provide insight to airlines or travel partners that a certain location has garnered a lot of interest and 2) that working with Airbnb to round out the travel experience my guarantee a final purchase (i.e., co-offering deals). Machine learning can help easily integrate those systems and track consumer search habits.
This was a very good essay. Very much appreciated the opportunity to learn more about how AirBnB is using machine learning to improve search. The one topic that I continue to think through coming out of reading this (and reflecting on the broader questions circling AirBnB) is the impact regulation will have on the business model. For many high-tourist cities, AirBnB has actually been a cause for concern as the influx of tourists enabled by the platform strains public services. Looking forward, I wonder if AirBnBs search algorithms can actually pick-up on customer vacation preferences to the extent that they can suggest less-trafficked destinations that fit the same archetype. In doing so, could AirBnB actually be on the leading edge of helping to better distribute tourist flows across many destinations and thus reduce tourist-local conflict?
Machine learning is proving to be a huge asset for companies like Airbnb because it offers an opportunity to scale faster and more importantly, create a good experience for those on the platform — the guests and hosts. I am fascinated by Airbnb because they are led by designers and find that even in solving technical challenges, also address experience design challenges. As a frequent user of the service, I am worried that with the continued growth of the company and more hosts coming onto the platform that there will be more opportunities to abuse the platform. More specifically, it makes it easier for property managers to list more and more properties, which of course, has greater implications for the housing market and local tourism.
Really interesting essay – thanks for the good research in putting this together! I think Airbnb is doing a lot of really interesting things with ML, but I would question the level of input data they actually have available to generate quality insights. To point, I would guess that most Airbnb users book 1-3 Airbnb’s per year, if that. How many data points does Airbnb need before they begin to extrapolate the user’s preferences? And, if they extrapolate incorrectly, from an insufficient data set, does that then hurt user retention? So I would just be careful of “over-optimizing” based on a small data set.
However, where this feels very powerful is in the optimization for hosts – there, there is a ton of data to leverage, and by helping their hosts succeed on the site, Airbnb sets themselves up for sustainable success.
I really enjoyed reading this essay. As a user of Airbnb, I appreciated learning about how they are making the experience that much more “friction-less” for both the users and the home owners and using machine learning to address these specific pain points. To your question on whether machine learning alone can propel the growth of Airbnb, I think the answer is that other factors are also important. Of course, the more users (and options) on the platform, the more valuable the marketplace. Another item that is very important is the entire user experience, some of which can be drastically improved by machine learning and some components might need a more intentional human interaction. I also wonder whether Airbnb is using visual search at all to scan and validate the quality of a listing – or if this could be a potential future improvement.
Great read with interesting insights! Being an AirBnb user over the years, I have sensed improvements in the search functionality. However, I never realized that the improvements were based on the connections between my searches. As I think about my experiences going forward, I really like the idea of the image classifications based on what I care about most. I assume this would be a major time save more me and would help me analyze options in a fashion that better suits what I am after. However, I am a little worried about how AirBnb might over-assume what I am after. Specifically, I have used AirBnb for very large groups as well as for two people. In each case, I am concerned about different things. How will AirBnb recognize the differences in what I am after in any given search?
Really interesting to hear how AirBnB is using machine learning, especially thinking ahead to photo classification and NLP to make the listings more relevant and useful to users. I’m a bit surprised of their use of personalized search given that there can’t be that many data points per user since booking a trip is an infrequent occurrence. I would also be concerned that the preferences could change significantly based on the trip (e.g., big group vs. romantic getaway). Similarly, for your recommendation to power search rank based on similar reviews, you first need the user to have written multiple reviews. I think your suggestion of expanding into the trip ideation process would be a great way to solve this issue because you’ll be able to collect more information earlier on about the type of trip and on the user.
Interesting piece on an app I have used on multiple occasions but did not realize the use of machine learning at different points. I do agree with the use of machine learning on the trip ideation phase. Their is a lot of power in the data set the Airbnb continues to accumulate, it could leverage this to introduce interesting new products that could be monetized eventually. One can look at Google as a prime example at the amount of products it has introduced, after having amassed a large amount of customer data – these included Flights, Images, Translate. Most of these are not monetized as yet but as the adoption grows I can see a number of ways in which Google will use them for its next phase of growth. Likewise using machine learning and predicative analytics I can see Airbnb trying out and introducing a number of new products that will allow it to growth into several adjacent business lines.
AirBnB is an interesting company to look at as they have been a big disruptor to the hotel industry. The question that I have is does machine learning help them bridge the gap between the reality of staying in a room vs what it appears like on the website. It is interesting that they are going to be using machine learning to read reviews and rank them. Also, while they have smart pricing for the hosts, do they also have a smart price alert for guests to let them know when prices are more favorable?
Really interesting read about how machine learning drive’s Airbnb’s growth and development. As I think about Airbnb’s goal to “own” the travel experience end to end, I do believe that machine learning can help them do this. However, for machine learning to be truly effective, Airbnb needs to own more data on the customer, upstream of the accommodation booking process and closer to trip ideation. Today hotel companies and airlines partner and share data back and forth to understand when customers are searching for or considering a trip in one destination. This allows the partner organization to launch targeted ads to capture that customer’s visit. Airbnb could consider similar partnerships to gain access to this data.
Interesting article! If AirBnB wants to own the whole travel ecosystem, curious to see how they plan to compete with big players like Expedia that typically manage the booking / transportation process and other large players moving toward the experiential aspects and also use ML / AI. AirBnB does a great job of have a very intuitive process for the end-consumer and agree their predictive capabilities will continue to differentiate them within their lodging niche.
So cool! I loved reading this. I always wondered how Airbnb did their price recommendations, and am impressed to learn that there are over 70 factors that they have incorporated into their algorithm. I often notice that different Airbnb listings come up for me vs for a friend when we are both searching for a place to stay, and now I know why! What fascinates me the most is the idea of potentially linking house preferences to Experience preferences and using that data to cross-sell. It would be interesting to see if people who prefer a certain type of accommodation (or even decor, as mentioned in your essay) would also gravitate towards similar Experiences.
Thanks for sharing this perspective and this brings up a point we have been discussing a lot in class, particularly around how using technology as a tool for helping humans make better decisions, instead of blindly relying on the technology to make decisions for us. That’s why i think the focus on natural language processing is super important because making sure the right reviews show up on the screen relative to what consumers are focused on is incredibly important. The more you know what a customer values, the more you can be prepared to address his/her needs when the time comes to book a travel. As it relates to your ideation point, if you have user data on past experiences, machine learning can help you reach the customer at the ideation process given it will be able to suggest travel locations, experiences and listings that will likely be enjoyed by the user. Super interesting!
Thank you for sharing this! The point about technology risks for ranking users, hosts, or otherwise encoding existing human bias into these algorithms seems pertinent given Airbnb’s particular history around this issue. In 2015, a widely circulated study from HBS found that guests with historically African-American names were less likely to be accepted for reservations compared to white guests, controlling for other factors (see https://www.fastcompany.com/3054520/airbnb-hosts-discriminate-against-black-renters-study-finds). Unfortunately, that discrimination can directly feed into the streams of data that inform pricing algorithm models or personalized search rankings (how many people have booked with you, the kinds of guests you’ve accepted in the past, etc). In 2016, AirBnb retained former Attorney General Eric Holder to create an anti-discrimination policy given these concerns, and Airbnb is still working to implement some of those recommendations. I would like to see the advances of these algorithms be complemented by a robust and ongoing assessment of indicator stats that signal how these online marketplaces are addressing discrimination risks (examples could include acceptance rates for guests by race, gender, or other factors; effects on submission rates when removing host photos from search results; encouraging instant booking, etc.).
You make a really good point on AirBnB trying to own the travel experience going forward. Machine learning and predictive analytics will be crucial for AirBnB to retain the customer and own the full travel lifecycle. However, I do think AirBnB will have to do more by way of partnerships with transportation companies such as airlines, bus services, online and offline travel agencies, etc.
I think the utilization of machine learning for AirBnB is very interesting. While I like the intention of using machine learning to analyze guest reviews, I think there is danger in using other guest reviews as training data. My concern is rooted in the fact that more often than not, extreme reviews (either positive or negative) are more likely to be posted vs the more accurate/average review. Therefore, I would not use crowdsourcing as a method of influencing algorithmic predictions. One interesting way AirBnB could also utilize machine learning is using learnt behaviors through user data to provide recommendations for actual trip destinations.