Welcome to Utrecht, the new HousingAnywhere Search Ranking Algorithm
Prioritizing fair prices and good behavior while hitting targets
As a two-sided marketplace, our ranking algorithm is arguably one of our most important pieces of tech. The way we develop it to decide which listings to show first and which tenants and advertisers to match really define us as a company. On top of that, it allows us to truly shape the impact we can have on people and society as a whole.
Last month, we released a new version of our ranking algorithm (HousingAnywhere Utrecht) that really moves us in the direction of where we want to go with HousingAnywhere: shaping the housing market towards providing higher quality accommodation, fair prices and a better booking experience. The previous version of our algorithm was based on a couple of solid assumptions: give lower scores to expensive listings and boost advertisers who already booked with us and proved their trustworthiness. These two were combined with a number of other factors related to the quality of the listing, such as a detailed description and the number of pictures. This resulted in a ranking which, a year ago, we believed best represented and assisted the dynamics of our marketplace. At that time, we used to work mainly with a few large Property Managers offering rooms.
During the last 12 months, though, we noticed that the algorithm’s behavior wasn’t 100% aligned with the vision we had for our company. Our number of users has grown rapidly, and, with it, the mix of searchers has also changed dramatically. In the past, we mainly catered to students between 20 and 25. Nowadays, the average age of our tenants has increased to well over 27 years of age. As we learned from our data, this group has different preferences when it comes to accommodation. The growing rate of apartments and studios booked on our platform provided a strong pull for our advertisers to list more of these properties, as well as for new ones to join our platform. This put even more pressure on our ranking system, as we were not just dealing with a growing number of listings but also increased diversity. We started noticing, for example, that apartments were ranked too low or that new advertisers were not getting sufficient visibility from our traffic, concentrating the demand toward only a subset of our inventory. None of these outcomes were desirable, either for the tenants or the advertisers.
One of the underlying factors we attributed this to was that the algorithm was, presumably, too simple for the complex inventory we feature on the platform. It was at that point that we started thinking about the next step: a brand new data-driven algorithm that would help us drive towards higher quality accommodation, fairer prices and a better booking experience — all while reaching our targets each month.
The HousingAnywhere Utrecht
Our new ranking (HA Utrecht) computes a score for each listing by comparing its price with average reference prices for similar listings in the same city. These values are updated once every quarter, in order to have our ranking aligned with the market’s fluctuations.
Internal research shows how users tend to stick to the results they see on the first page and hardly look further — as you know, the best place to hide a dead body is on page two of Google.
The basic idea of our new algorithm is quite simple: we want to give our tenants the best we have to offer whenever they visit our platform. As in every ranking algorithm, the concept of what is the best you have is what actually defines you as a marketplace. Internal research shows how users tend to stick to the results they see on the first page and hardly look further — as you know, the best place to hide a dead body is on page two of Google. Less than 10% of our visitors go to the second page, and just 3% take the risk to go on to the third one. Both our data-driven culture and our habit of working with data-based products have been crucial.
The early assumption was that the market behaves pretty much in the same way regardless of the city, but our data suggested very different patterns according to the region. Our previous approach to search, though, was too static, and it didn’t allow for the flexibility to fine-tune it according to the dynamics of the city. We decided to not only revise our current ranking parameters but also reconsider the set of factors that define good market behavior. In that way, we could make the market flourish.
The core of what makes a listing attractive is, among other things, meaningful pictures and a detailed description. But being an international platform where different listings are created 24/7, we noticed something else: the concept of what is good differs from city to city. The obvious one is the price differences: a listing considered affordable in London would be considered incredibly high in Berlin.
Less obvious ones also come into play, like when prices for student rooms are going crazy while rental prices for apartments remain stable. We had to deal with all of this.
We learned that determining a good price for a listing can be based on the wrong expectations or incomplete knowledge of the market dynamics, which leads to lower-than-expected returns and general disappointment.
Whether you are a professional advertiser like a property manager or a private individual managing your family’s assets, price is the single most important factor in defining whether you will be successful or not — that is, whether you’ll rent out your listings and generate the anticipated income. We learned that determining a good price for a listing can be based on the wrong expectations or incomplete knowledge of the market dynamics, which leads to lower-than-expected returns and general disappointment. In order to help our advertisers be successful on our platform, we concluded that we needed to better align with the market starting at the ranking algorithm, while at the same time offering better information and guidance in setting the price (more on this in a new blog).
But this is just the basics, because once you created the listing, then you have to take care of it. You have to reply to messages, and you have to be quick in doing it, ideally less than 5 hours for 95% of the messages you’re receiving. It’s our vision as a platform to prioritize listings based on the quality and good behavior of the advertiser, for all of the listings that our advertisers are creating.
Better performance and a better price
Since HA Utrecht has been operational, we’ve increased the number of times users click on listings (CTR) by 8% and, equally important: the average price of listings shown on the first page has reduced by 40 euros, and the quality of the mix of accommodation on offer has greatly improved. Overall, we achieved to offer an improved user experience by showing a better looking and better mixed search result page.
We feel that the new ranking is bringing us closer to fulfilling the needs of our tenants, while also enabling us to have the impact we want on people and society. The surge in rent prices across European cities in recent years has heated up the debate on the affordability of housing. This is particularly affecting the younger segments of the urban population, like students and young professionals. The trade-off regulators are facing is: how to sustain the international mobility of human capital (by keeping housing affordability in check) whilst allowing individual and professional investors to generate returns.
At HousingAnywhere, we feel that we are in a unique position to help the market find that balance and, unlike regulators, can leverage our data and platform to assist the market agents toward price formation. Our goal as a marketplace is to maximize the volume of the market by facilitating as many transactions as possible and that can ultimately only be achieved when both sides perceive the price as fair.
We Evolve: Transparency at the next level
We are happy, but not satisfied, so we keep iterating on the algorithm! Every month we run a new A/B test that aims to improve HA Utrecht. Our latest update allows us to show bills and additional fees in the formula to be as transparent as possible in the pricing and minimize any possible surprises for our users. We are working hard to take our algorithm to the next level, considering more granular reference prices, ideally to a neighborhood level. Each city will be meticulously monitored by collecting up to 50 different averages that will be compared to the listing’s price and injected in our formulas to give an even better estimation of the actual listings’ quality.
Now, what does the future hold for HousingAnywhere? Our user research shows that our tenants are more likely to book with our platform again if they identify with our brand values. We see the new ranking algo as a major step toward our goal of becoming the platform to actually manage and live the renting process end-to-end: through multiple rental contracts, across different homes and cities.
This article was written in collaboration with Massimo Belloni & Gianluca Valentini.
