AirBnb: A path to a succesful business

The platform makes it seem easy to rent your listing. However, you may have to pay attention to a few details if you plan to grow a business out of it. Dive in to discover how to strategically make profit with the platfrom.

From being a simple idea born inside the minds of two men, Brian Chesky and Joe Gebbia, unable to afford their rent, to becoming one of the best vacation rental sites, AirBnb has come a long way. As a matter of fact, the “air” in AirBnb came as a result of Brian and Joe renting an air mattress on their living room floor.

A Bit of Background

Since its foundation in 2008, the company experienced a continuous growth with a boom after 2012, ending up to a revenue of 2.6 billions US dollars in 2017, 12736 employees in 2019 and an astonishing 2 millions people lodging with AirBnb each night in October 2019. Today, AirBnb offers listings in 191 countries and counts about 150 millions users.

Here-above is a graph depicting the worldwide presence of AirBnb from data obtained through Inside Airbnb. The bubbles sizes are proportional to the number of listings in each city. In total, approximatively 1.5 millions listings are present in the dataset!

With such astonishing numbers, one cannot help but wonder how to get a piece of the pie. More specifically, how can short term renting be turned into a sucessful business, to the point of being a source of income? If you are interested in the idea of subletting your properties and are considering AirBnb as the plateform to do so, then this data story is for you! Indeed, with the help of a thorough analysis of AirBnb data, this article will try to solve the mystery of profitable rentings by determining what makes a listing sucessful.

Thus, we seek to find answers to the following underlying questions of our problem:

  • What do people look for when booking an AirBnb ?
  • What variables have the largest impact on a listing’s success ?
  • Are they the same in every city ?

Defining Success

Before getting down to business by pointing how to get make your listing reach the top, let’s get everybody on the same page by detailing what it means for a rented property to be successful. Quite naturally, the goal when renting a vacation housing would be to get the highest rating accompanied with good reviews and have custumers all year round. There are other indicators such as having customers come back to your place but for the purpose of this study only the three factors previously mentioned will be taken into account. Since customers rate a listing over several criteria such as cleanliness or communication, an overall rating that includes all of these aspects will account for client satisfaction. This is the role of the review_scores_rating metric. Due to AirBnb’s international presence, reviews are found in many languages, with English being the most common one. From the latter, sentiment analysis is carried out to evaluate both the overall positivity and overall negativity of the reviews. These two measures are summarized in a so-called ‘compound’ measure that defines the global sentiment of a comment. A compound value greater than 0.05 indicates positivity whereas values lower than -0.05 are related to negativity. For values in-between, reviews are considered as neutral. The booking frequency, our third indicator of success, is estimated with the average number of reviews given to a listing every month. This is under the assumption that, in average, there is a similar proportion of customers leaving a review of their stay. This is only an estimation since not all guests leave reviews. In addition, the duration of the stays could vary considerably. It can nevertheless be stated that this metric provides information on the demand for a listing.

Now that we can all agree on a common definition of success, we can move along to getting an understanding of what parameters have the most impact on a listing’s so-called success. To do so, we developed a complete pipeline integrating some Machine Learning algorithms to perform a preliminary exploratory analysis.

Zooming into a city at a time

As a first approach, the analysis aimed at understanding the parameters that influence the most the success of an AirBnb listing in Amsterdam. Focusing on Amsterdam only allowed us to strategically build an efficient framework that would later be applied to other cities, as we hypothesise that some parameters may have location-dependent importances. This city was selected as a reference to develop the analysis workflow as it ranks 25 in the most visited cities worldwide, and lies in the top 10 most visited european destinations. The elevated touristic flux, concurrent with the decent amount of AirBnb listings contained in this city’s dataset motivated this choice.

Working with this single city, a highly efficient and generalized framework was constructed, reducing the computational cost in comparison to building it with the worldwide dataset.

Based on age, interest, financial means, country of origin and so on, the appeal for a given listing may be very different. Though there might be a high level of variability in the data, we still expect to manage to identify certain trends in terms of which features contribute most to which success metric. It is most likely for anyone who has ever gone on a touristic trip to develop preferences in terms of housing to some extent, whether for the price range, the distance to touristics attractions and transportations or the type of accommodation. While most of this information is present in the Inside Airbnb dataset, it lacks any clear reference to the proximity to public transportation! For this reason, we integrated supplementary data with the open-source citilines database. As not all the corresponding cities are available in this database, we reduced our initial dataset to contain only cities for which the public transportations data is available.

Before diving any further into the analysis, we will shortly describe the implemented Machine Learning approaches that were used. We selected Random Forest (RF) models because of their high capacity in generating high-performance and computationally efficient models. Standardisation as well as hyperparameter tuning and multiple performance assessment methods were considered. Machine Learning was also applied for the natural language analysis of comments guests left to listings they stayed in.

With these tools, we can start to answer to our research question. The importance of all the features in the Random Forest predictions was determined for each success metric: the listing’s rating, the mean sentiment of its reviews and the monthly rate of reviews. Sankey diagrams allow to visually summarize the results. We provide a dropdown menu allowing to compare different cities. The thickness of the connection between a given feature and a success metric is proportional to the contribution of this feature to the prediction in the RF model.

The Sankey diagrams were obtained by setting a threshold such that only links of importance higher than 5% are displayed, such that only the highest contributions are displayed. A few conclusions can be drawn from observing these graphs:

  • The distance of a listing to transportation (dist_nearest_station) seems to be a determining factor in most cities.
  • Some slight differences can be noticed: the rate of host response appears to be a more important factor in determining the amount of reviews per month for Berlin than for Manchester.
  • Overall, similar features, such as number_of_amenities, dist_nearest_station, price, host_since,… seem to appear for all cities, though often contributing with different weights to the success metrics.

So say you are interested in launching an accommodation business on the plateform in the city of your choice, for example Barcelona. You will be able to deduce from the Sankey diagrams that you should pay particular attention to the number of amenities you offer, the price you ask, as well as to whether transportation is in close reach. Moreover, you will notice that the quality of your offer is not the only factor to influence the success it will have. In fact, guests also look into the host behind the listing. As a host you should keep in mind that your experience (host_since, host_total_listings_count) will greatly impact the success of your listing.

Up to this point, the analysis was focused only on single cities, independently from the results other cities obtain. Therefore, to get further insight and to be able to generalize, we provide a globalised analysis, trying to identify similarities between different locations.

From single cities to a worldwide view

We previously established that for different cities, different factors are determinant of a listing’s success. It would thus be interesting to see whether the AirBnb platforms of different sets of cities can be clustered according to their similarity in the most important succes-determining features. In doing so, we may be able to infer what are the common characteristics of these groups of cities. It would also be interesting to determine which features are conserved throughout the different clusters.

To compute the similarity between cities, only the list of the top 5 most important features are considered. Extensive research was performed to identify the best algorithm and Rank-biased overlap (RBO) was finally selected, as it has interesting properties such as non-conjointness groups handling (different elements in the compared lists) and the highest weighting of higher ranks. The clustering effect can be visualized in the heatmap below. The colorscale indicates the degree of similarity between feature ranking for different cities. Clusters can easily be identified as diagonal light boxes. This analysis was performed for all of the success metrics as target through the RF model, as well as for a multi-target component, which consists of a simultaneous combination of all the success metrics.

Depending on the selected target, a clustering effect can be visualised. Clustering is defined by a shared set of relevant features. This phenomenon can be observed for three out of the four targets: the number of reviews per month, the overall rating and also for the combination of success metrics (i.e. multi-target). The fourth target, that is, the sentiment of the reviews, shows homogenous levels of similarity throughout cities. Clusters can still be extracted from these levels of similarity, but it should be noted that there will be little to no difference between clusters for this metric. This peculiarity could be interpreted as being due to the fact that guests tend to base their reviews on similar aspects, regardless of the country of the accommodation.

In order to get further insight on these clusters of cities, we created bubble scatter plots (see below), which show the features that define a cluster of cities and their ranking of importance in the cluster.

These bubble scatter plots confirm the previous observations made on the heatmaps, as features are ranked in the similar manner for most cluster. Moreover, we observe that for a single metric, the two top ranked features are quite similar between the clusters, meaning that the features discriminating the clusters are rather bottom ranked features. Concerning the multi-target analysis, the feature ‘dist nearest station’ is of prime importance for six clusters out of seven! The other two features that stand out are ‘host_since’ and ‘number of amenities’. Therefore, for you as a host, a good suggestion would be to invest your time in finding a place to rent that is close to a public transport stations and that offers sufficient (and of good quality) amenities. Finally, we suggest you to stay patient, as the longest you have been a host, the more successful your listings will be!

However, we cannot identify characteristics specific to each cluster because we would probably additional information that our current dataset does not provide us, distance to touristic attractions (center, parks, beach,…). If these aspects could be included, we would most likely be able infer the common characteristics that describe the cities that form a given cluster. For example: summer destinations, cities with many indoor/outdoor activities, …

Conclusion and discussion

At the end of this analysis, we truly hope you will get the best results (and income) with your listing. Even though the competition is tough, remember that few elements can improve tremendously your chances of success. Also keep in mind that the parameters that most influence your future success will also depend on where your housing will be located. However, tourists’ expectations are conserved regardless of the destination. As might have been expected, most travelers aim for listings close to public transportations (distance_nearest_station) rented by a host they feel they can trust based on their experience (host_since) and that has a lot to offer (number_of_amenities).

All the images are open sources and retrieved from pixabay.