Shaking Up The Housing Market With AI & ML
This story explores issues in the real estate market. Digitisation and emerging technology like artificial intelligence open up exciting opportunities to meet these challenges.
Spending the last days digging deep into recent changes within the housing market was depressing. Headlines like these are evidence of that; “The Vanishing Affordable Apartment”, “Unsheltered”, and “The Rental Market is Broken”. Housing costs have risen dramatically — not only in most parts of Europe but also in the US; large cities are especially hard hit by this. Since the start of the new decade — shortly after the burst of the housing bubble — to rent or to buy a home has become less and less affordable, even though wages grew. Since 2011, Mike Rosenberg reported, “rent has grown four times faster than incomes, and home prices have shot up more than five times faster than pay.” The results of this for most of the global bigger cities, like Zurich, Munich, Berlin, London or Los Angeles are: Housing prices have skyrocketed and those who can’t afford them struggle. The consequences of this development have hit individuals and businesses (e.g. architects, developers) alike. There are different aspects that are causing and/or amplifying the problem:
1. Search tools are outdated
Most of the search processes are run via property portals where landlords offer up their properties to attract prospective clients. But there are several problems with that. In order to market their properties the best light they put up pictures — often staged or heavily modified — and provide other data that is difficult to verify. In addition to that the filters for the search are more or less insignificant; based on a few features like size or availability of balcony or elevator. They don’t differentiate amongst features — not all balconies are nice balconies. Nor do they provide any analysis to make a reliable statement about the quality of the building or interior facilities. When you are lucky enough to come across a promising offer you end up discovering that it’s no longer available. This makes the current search process time-consuming and unreliable in terms of the “real” qualities of the property.
2. Lack of information for decision-making
The emptiness of the information currently provided by property search tools leads to poor and uninformed decision-making. The lack of real-time analytical tools also hinders the ability to carry out accurate valuation or to calculate investments and debts. Additionally, most of the sensitive information like rental values are kept behind closed doors by brokers and real estate agents.
3. Insufficient valuation tools
All the aforementioned aspects play into a simple fact. When real estate is not assessed correctly but is instead subject to deficient information this leads to wrong valuations and paying skyrocketing prices for properties that would simply not be worth the money — given the full picture.
Using AI and ML to Improve the Housing Market
Especially with the current price dynamics, it is important to prepare real estate decisions well and not to pay usury purchase prices and rents. (Dr. Nima Mehrafshan)
New developments in the so-called PropTech market bring a lot of new tools to make property search, assessment, and valuation smarter and more efficient. Artificial intelligence and machine learning are among the most highly touted types of PropTech. Some of those PropTech companies aim to address the three issues discussed. Aiming to make processes like spacial analysis and valuation faster and easier; in the best case leading to valuations (and prices) that are fairer than they are today.
Until Zillow — a company analysing thousands of properties and generating price estimates — came on the market in 2006 tools for measuring the qualities of architecture were crude and unreliable. But Zillow’s so called “zestimations” are insufficient. Although it is one of the largest real estate portals in the United States, and claims to predict prices within 5% accuracy roughly half of the time it still has to be considered that those price estimates contribute to setting prices in the first place.
Recently, Zillow advertised a new feature which is supposed to make the estimates of the properties even more precise. The company uses artificial intelligence which recognise different types of rooms such as kitchens, bathrooms, bedrooms on housing photos. It then clusters and rates them in order to determine the standard of housing. According to Zillow this helped to make the home price estimates about 15% more accurate. But taking a closer look reveals how blind the process still is. Certain information such as the number of square metres and price levels in the neighbourhood are quite easy to gather, and also the approach of photo analysis can help. But at the moment, given the fact that this feature is still in its early stages it seems that they are still far away from being able to capture the qualitative differences which are crucial for decision making.
Things that humans can immediately grasp with their perception, such as the atmosphere of a place, the proportions, the light conditions, can not yet be determined by artificial intelligence and, above all, do not play a significant role in the evaluation of companies like Zillow yet.
However, one could get far more out of the available data than has been possible so far. Archilyse has taken on this task by bringing together available floor plan with urban contextual data to capture subjective qualities of a space in order to provide users with better tools for predicting the final value of a property. With geo-referenced floor plan data Archilyse can not only determine how well a floor plan meets the specific needs of someone, but also simulates actual living conditions. For example, it calculates the amount of sunlight in an apartment during a day and the views of certain points in a space. Determining these aspects was previously reserved only for human judgment, which is now based on computer-aided calculations.
“Teaching computer the same sensibility that a person has, can help the searchers to find a good flat.” (Michael Crook)
Furthermore Archilyse matches the preferences with individual points of interests, like distances to workplaces or child care facilities. Pairing these analytical results with price estimates can help to include subjective qualities to make predictions more accurate. This analysis now makes knowledge for decision making available to laymen too.
“An objective property valuation is a highly complex process. […] [A]n elevator for a higher unit of an apartment building means an increase in value. For the apartment on the ground floor, however, it has mainly cost disadvantages.” (Dr. Nima Mehrafshan)
Similarly, more complex interactions between a variety of real estate features need to be considered. This is simply not possible without modern data analytics.
Unlike Zillow, Archilyse does not merely use the most easily quantified information, it uses precisely the information that the brute force approach employed by big data models neglects, and attempts to weight it accordingly.
Obviously, for all purchase or rental decisions minimising the costs is the primary goal. This is the case for landlords, owners, developers, and renters and purchasers alike. Said in other words maximising space is in the focus. But if it could be understood as maximising the quality of space development might take on a rather different aspect. In other words, quality over quantity.
It is here that machine learning comes in. It can provide tools by which rigorous predictions regarding the quality of space might be made. If ceiling heights can be proven to matter; if careful modulation of space can be shown to work; if natural light makes money — developers will pay attention to them.
These tools can help to change the approach to architectural qualities and put into focus human-serving design; not profit. Adding this analytical knowledge to the market for rental or purchase prices and making it accessible to everyone can have a positive impact. While rents that are reasonable for tenants are currently also horrendous, this newly available knowledge could lower prices. This, eventually, serves the market to become more transparent and fairer.
If you are interested in Architecture Analysis check out other articles on Medium or follow us on Twitter.
