From self driving cars to better online advertising, to composing music to diagnosing cancers – ‘machine learning’ has seemingly infinite applications. Artificially intelligent chess robots being the least impressive of them.
Machine learning, in a nutshell, is the art of teaching computers how to teach themselves. Traditionally, computer programs can perform tasks because their code has predefined all the parameters and data needed – the computer just runs through the motions it’s given. But what about problems that don’t have definite answers? How can we teach computers to, say, caption a previously undefined picture or drive a car through uncertain traffic?
Machine learning under the hood
Machine learning algorithms let computer programs improve the way they perform a task based on pattern recognition and past experience. In the same way as I’ve recognised that food stops magically appearing in the fridge once you leave home – computers can now better themselves without someone having to spell it out for them. By feeding an artificial intelligence with a tonne of data, asking it to make predictions, and then rewarding it for getting things right, over time it can better its performance. It ‘learns’ the best way to solve a problem through trial and error.
This isn’t new technology. We’ve had rudimentary self learning machines for years now. What’s changed, however, is the increasing volume of information we track and the new openness of previously closed data sources. (Not to mention the heightened sophistication of machine learning models thanks to some pretty clever developers). Out of nowhere, machine learning software has suddenly become practical.
Autonomous vehicles, for example, can be taught to use data to drive themselves. They can identify objects (stop signs, pedestrians, road obstructions, other cars on the road), track speeds and view traffic reports to get people from A to B without any unfortunate consequences – even if they’ve never taken that route before. Facebook uses its enormous database to recognise people in photos and automatically tag them. Google takes a user’s interests and search history into account to deliver them more and more relevant search results.
Now it’s real estate’s turn. Machine learning has well and truly entered our industry. So what will this new technology mean for us?
Real estate data sources
Machine learning is heavily dependant on the information you feed it. The more valuable data you can put in one end, the better the output is at the other. Let’s take a look at some of the data real estate software developers have to work with:
- Infrastructure: water, sewerage, power, gas and telco – all publically accessible and highly relevant.
- Demographics: wages, ages, interests, gender, etc. Vendor and buyer information is going to be hugely valuable for prospecting and buyer matching applications.
- Traffic: it’s not just self driving cars that will benefit. Consider commercial real estate applications for this sort of data.
- House prices and sales: self explanatory, really. Cost market analysis and market predictions will greatly rely on these data sources.
- Upcoming developments: planned future developments are going to play a big part in predicting the hot spots of the future, enabling algorithms to catch gentrification before it happens.
- Current listings: prices, visual data, location and key details. Matching a buyer with their perfect home might be a job for machine learning in the future.
- Buyer online behaviour: in relation to the above, profiling buyers themselves is also very important when looking for the most relevant listing results.
- Client data from CRMs: while this isn’t public information, certain machine learning systems might allow for you to plug your own data in for even more tailored results.
- Website analytics: traffic, bounce rates and demographics – this will be increasingly important when predicting which demographics are showing interest in buying a home, and which areas are increasing or decreasing in popularity.
- Social media: similar to website data, but with far more personalised results. Social media allows for things such as ‘lookalike audiences’, which can greatly increase the reach of your marketing and prospecting activities.
That’s just the beginning. We’ve got volumes of highly relevant data to go on. Being able to utilise this en-masse with clever software opens up a massive opportunity. We meagre humans simply can’t hope to garner any bite-sized, actionable insights from data-sets the size of the ones we have. After all, how are we supposed to see the big picture when the picture is basically infinite?
Thankfully, computers can now take over all that pesky pattern recognition for us. And to great effect.
Potential effects and applications of machine learning in real estate
First up – there are now some pretty nifty machine learning methods of predicting the future of the property market.
Accurately evaluating property prices is extremely important for real estate, the stock market, tax sector, economy and size of buyer and sellers’ wallets. While we’re reasonably spot-on with our current predictions, we’re pretty limited by the scope of data our current systems can take into account.
Normally, pricing is done via pretty basic comparable market analysis. Similar houses in similar areas are used to give an estimated price for a given home. But what about other factors? Nearby schools? Police and fire stations? Planned developments? Recent interest in the area based on portal website traffic? Walk scores?
Creating systems that can do this for us automatically is big business, and plenty of companies are already taking the challenge on board. CoreLogic, for example, have been hiring software engineers with machine learning experience – and we’re bound to see the fruits of their labours in a big way soon enough.
This technology is going to be massive for the industry. Pricing and growth predictions on certain areas might be able to help us identify the next booming suburb – a huge benefit for investors and property developers – or simply get your vendor a better price.
Property development and identifying economic corridors
Taking market watching even further – higher level machine learning insights will have a huge effect on property developers, urban planners and government.
Giving greyfield areas a shot in the arm is a massive problem for cities around the globe. Usually funded by taxpayer dollars, you want to ensure the money you’re spending is going to actually catalyse some sort of economic growth. So how can we pinpoint the best way to stimulate these dead areas? Surprise, surprise – this is a job for machine learning.
The advent of open data sources for things like soil quality, pollution, economic status of certain areas and zoning will soon play a massive role in the development of the urban sprawl. Plus, governments will be able to easily identify candidate areas for low income housing developments without having to build from scratch.
Likewise, we’ll be able to gain much better insights on greenfield farm or salvageable brownfield industrial areas – encouraging governments to build or update transport routes and other infrastructure to encourage more production.
Plus, finding the most effective economic corridors between industrial/commercial areas and suburbs with dilapidated housing might encourage more investors, buyers and property developers to buy up and redevelop the cheap real estate. The rebuilding process could be paid for not by tax dollars, but through capitalist incentive.
Here’s a fun one. Give some machine learning algorithms enough conversations to read and they’ll start figuring out the rules of basic chit-chat. This is exactly the idea behind ‘chatbots’: little programs that integrate with messenger applications that can – believe it or not – hold a conversation with you all on their own.
While chatbots are still in their early days (and ergo have pretty limited scope) they’re already able to automate a lot of the smaller administrative, customer-facing tasks. Plus, they can do it 24/7.
Imagine, for example, a chatbot integration on your website? If someone wants to know how much commission your agency charges, for example, the chatbot can answer. What’s more, your chatbot would be able to ask leading questions to create rudimentary profiles for agent’s further use: “Hi, welcome to example.com – are you looking to buy or sell?” This information can be sent directly to your CRM as a lead, with the initial nurturing taken care of for you.
We’ve barely even scratched the surface on their potential effect on our industry in this article. For a much more comprehensive discussion of chatbots’ effect on real estate, see our blog.
Like Google, a portal’s job is to provide the most relevant results for any given query. Though, unlike Google, they have limited data to work with. Google uses user profiles to determine your interests, rank results based on overall relevancy to an individual, and more. But portals generally haven’t made the most out of those resources (yet). If someone’s looking for a three bedroom house on realestate.com.au, they’re only really able to show all their three bedroom houses in order of recency. How much more tailored can the results be?
With machine learning, computers can get to know you. Frighteningly well.
As Robert Scoble said in his book, The Age of Context; “The more technology knows about your, the more benefits you will receive“.Reasonably Orwellian, yes – but true nonetheless.
What if a user had clicked on all the houses with marble kitchen tabletops? Or what if they’d showed an affinity for a Jacaranda tree out the front? If portals adopted object recognition and machine learning algorithms, they’d be able to deliver far better results to their potential buyers. It wouldn’t just be the price range, location and number of bedrooms taken into account – it could be all sorts of fine-grained data. Without the manual data entry. Show an interest in terracotta roofing tiles, and your favourite portal might be able to suggest similar listings – much like how Amazon will show you relevant products based on your previous behaviour on their website.
Machine learning gives portals the ability to deliver your listings to the most qualified leads on a per-listing basis, making it harder for your properties to get buried under the slew of new ones. Ergo, faster sales and better pricing results.
Okay, this is where it really starts to get cool. Meet Ross: an artificially intelligent lawyer.
Well, actually, Ross is more of an assistant to lawyers, helping them get more done by researching faster than even the most seasoned librarian could, and thereby leaving lawyers with more time to focus on advising clients.
But who knows? Perhaps a version of Ross might even be released exclusively for real estate in the future? Real estate is rife with legal issues, and agents’ and administration staff’s time is constantly being wasted nose deep in some heavy legal tome or fine-print website to ensure a contract is up to snuff.
Artificial intelligences like Ross will have much further-reaching effects in the near future. Look forward to it.
Machine learning will boost productivity
“Give it to me straight, Doc.” You say. “What does all this mean for agents?”
At the end of the day, any fears that your role will be eventually taken over by some SkyNet, real estate terminator are reasonably unfounded. Sure, I have no idea what’s going on in Google’s R&D lab, but one thing I do know is that after all my research into machine learning’s effects on real estate, it so far has only served to make agents better at their jobs.
Whether it be machines keeping an eye on the market to warn you of peaks and troughs, chatbots nurturing your leads, or portals delivering a better service to ensure your listings go to the right buyers – machine learning is only going to save you time and headaches.
Machine learning and AI are here to assist us. Not steal our jobs. These are extremely exciting times in tech and real estate. Bring on the robots.