Are Rightmove and Zoopla still disruptors, or soon-to-be disrupted?

Computer Vision and Machine Learning will divide by ten the time it takes to find your next dream home.

Understanding properties features at scale make a deep-discovery of the stock a much easier process.

It is ten o'clock on a week night. After a long day at work, you finally find the time to go to Rightmove or Zoopla and resume the search you started ten days ago: finding the next house or flat you'll buy or rent.

And like every single time, you receive back dozens of search results. Despite entering the same details over and over, theses portals have zero memory, zero understanding of what you are looking for. They fundamentally are blind to your real preferences. Are you looking for an open space layout or rather small, cosy, rooms? The alerts you receive on a daily basis are not any better. They are a mix of period properties, ex-council houses, CGI-ed new builds.

If this sounds familiar, you are not alone.

During 2018, we will collectively spend 150,000 working years on the property portals, at a cost of £5bn to the UK economy.

Incredibly, the time on platform is a metric than the portals follow and report: last year we spent an extra 40 million minutes on Rightmove, despite a falling number of transactions. These platforms behave like cab companies extremely proud to make you wait longer. However, more time spent does not translate into higher revenues for the real-estate agents, nor does it highlight a better experience for the users. We believe you could find what you are looking for in a few clicks, rather than hundreds.

How? With Computer Vision and Machine Learning. Even if you are not familiar with Computer Vision, you have been already using services and apps leveraging it: when clicking 'images' after a google search for instance or when Facebook recognises your friends in a picture.

Our model at work: automatic features detection is one of the cornerstone of the property portals of the future.During the past five years Computer Vision technologies have been doing extraordinary progress. NVIDIA, a leading manufacturer of deep learning processors reports that the performance improved by a ratio of one to 500 during the past 5 years. And the rate of improvement is not slowing: one to 10 over the past six months. Off-the-shelf APIs already exists (Computer Vision as a service, if you want) but for the moment they can still only recognise simple concepts in images. This is why we have been training our own models. For instance, we are able to identify period or modern features, windows sizes and ceiling height. More fundamentally, this enables us to approach property search at a semantic level: understanding what makes a specific property attractive to you.

As for Machine Learning, it is even simpler. Recommender Systems have been already successfully applied to music (Spotify) and movies discovery (Netflix). If they have not been applied yet to property search it was for the simple reason that it was not possible, until now, to effectively automatise the features identification process. Imagine a property portal where each picture you like and each property you follow produce information about your preferences. Our algorithms use theses inputs to match you with the right properties. The search can then be refined by commuting time, and the match between you and the neighbourhoods.

Machine Learning: how to develop personalised properties recommendations

Love a yellow bathroom? It should not take more than half a second to find one.

Finding your perfect property might soon become as simple as connecting your Pinterest board and entering your and your partner office postcodes.

Or to get the results of really complex search: find me all the loft with apparent bricks, less than a 30 minutes commute to my office, at less that £850 a square feet. Theses are the solutions we have been working on during the past twelve months: building the next generation of property portal.

Finally, at a higher level, we found it is absurd that a pointless empty box built with no details, has the same visibility that an attentively designed property. The blindness of the property portals might have contributed to the deterioration of the new builds quality during the past fifteen years. By lowering the visibility of insignificant goods, property search in the future will penalise bad products. Using technology to enable users to express their true preferences will support the production of higher-quality architecture, by more efficiently matching theses products with the people desiring them.

Will we be the ones building the solution of the future? I don't know. But it's worth trying. And we don't mind floating theses ideas openly, as we believe it's more important than this problem is tackled sooner than later.

If you want to share ideas, discuss the next-generation of property portals or simply the proptech sector, email me at daniel@cortexai.io

Stay up to date with the latest PropTech trends and news

Receive our weekly newsletter every Tuesday morning.

Berlin

5th September 2019

von Greifswald, Berlin

Buy tickets

Vienna

28th November 2018

Sofiensäle, Vienna, Austria

Buy tickets