The Neural Real Estate

January 25, 2021 by R&D Bonumic

Share to  

Facebook

Twitter

Linkedin

Get link

Share to  

Facebook

Twitter

Linkedin

Get link

Well before the industrial revolution, change and disruption have required workers and businesses to adapt to emerging technologies. Today, AI is transforming the real estate industry, allowing commercial real estate firms and local governments to leverage the power of AI to reduce costs, increase productivity, and boost revenues.

What is AI?

In a very broad sense, AI (artificial intelligence) is the technology that can develop logical conclusions the same way a human would – or sometimes even better. AI employs complex, highly developed algorithms, and sophisticated analytics to predict future outcomes or behaviors.

Everyday examples of artificial intelligence

Today, it’s hard to imagine what life would be like without a computer or smartphone. While most people don’t realize it, AI is already being used ‘behind the scenes’ in many parts of our everyday lives:

Why AI and Real Estate Are the Perfect Match

In the whitepaper Real Estate Predictions 2020: Conversational AI, global professional services firm Deloitte calls artificial intelligence a potential game changer for the real estate industry.

Source: Deloitte Center for Financial Services analysis

The firm notes real estate is facing significant disruption – and potential opportunities for improvement – as AI changes the way people work and live. As property values continue to rise and business becomes more competitive, stakeholders such as investors and tenants are understandably becoming more and more demanding.

In response, real estate practitioners have begun leveraging AI to boost productivity, decrease costs, and minimize manual errors, all of which helps to increase profits for the real estate brokerage and its clients.

In 2018, there was the first AI-driven real estate transaction, acquiring two multi-family buildings in Philadelphia for $26 million. This property was picked by the so-called “soon to market detection” algorithm that defined whether it was going to go to the market.

This was a result of analyzing tens of thousands of data points to define interesting data such as:

Yes, this already happened and it’s just a beginning. (source)

As real estate firms incorporate artificial intelligence, repetitive administrative tasks begin to decline, creating more time for work that can directly and positively grow revenues and increase the bottom line.

More than our brains can handle

In the real estate buying and selling business, at the very basic level, we think of data in just a few critical data points: location, price, and volume. If we think deeper, we can find ourselves creating a data model where we involve parameters like property amenities, avg. price in the market, price history, supply, and others.

However, we are nowhere near capable of taking thousands of data points into account when making a decision on our investment, which means we are not taking nearly as close to everything into account. However, our machines are capable of performing such tasks, and every day they are more capable and with every successful prediction – they are better.

The value of using AI in the real estate business isn’t only if we can make improvements, but if we can make new possibilities.

Nowadays, being able to analyze thousands of data points in a fragment of a second and make a prediction — is what makes AI technology unmatchable to any solution in the real estate market.

The challenge today isn’t what we can make out of all the data, that is on its natural road of progression, but the bigger challenge is – how do we gather the data necessary to train and improve the AI technology.

The sources of real estate data are infinitive and rarely come in a format that professionals can use out of the box. Without being able to collect and aggregate data at high speed, it will be challenging to produce better results at almost any advanced AI solution, but also it will leave us with repetitive and manual processes of inputting and updating.

Real Estate data on autopilot

Before analyzing and acting upon data – gathering that data is one of the most tedious tasks in real estate, having to enter and update the same property data over and over again. Real estate brokerages use many sources to track properties on the market and no matter where they store the data (spreadsheet, MLS, CRM), having to input and update manually is a painful process.

Instead, imagine if all your real estate offering emails and documents are daily scanned, organized, and listed without you even moving a finger — letting AI technology automatically update your property database while you do much more profitable things like previewing property, meeting with clients, and doing deals.

AI model to extract real estate data from documents and images

We have talked to real estate brokerage companies and decided to create a machine learning model that will turn real estate flyers into searchable data:

Digital real estate flyers are one of the common ways of sharing listing information with clients, prospects, and other brokers. The challenge is that gathering them and manually entering information from a real estate flyer is labor-intensive and very expensive. Brokerage companies, depending on the size, spend 2-3/h a day extracting data from emails, flyers, contracts, etc.

The good news is – because real estate flyers are actually free form digital marketing materials in a PDF or image type format, we can use AI technology to extract and upload key information from every page of the flyer such as floorplans, satellite maps, pricing, square footage, address, broker contact information and much more.

We took the challenge of creating the first AI step to extract, translate and understand real estate data from any given image or document.


Computers don’t see images the same way we do — they pretty much understand images only as a file format, their size, their location, and others. Therefore we had to teach and train the machine to be able to read and organize data the same way we do, but to do it 100 times faster.

Unlike typical texts in articles, books, and so on, which are usually completely sequential and where sentences are related to each other, a large amount of text in the flyers is scattered and unrelated with each other.

Using multiple computing processes like OCR (Optical Character Recognition), we have extracted human-readable text entities from 1000+ flyer images and PDFs to create a sizeable raw text material of real estate data that our models can use to learn more about property features.

Building this computing process will allow us to extract data from almost any file format in the future, as long as there is a visual representation of the data.

Nature of the texts in the real estate flyers – unlike typical texts in articles, books, and so on, which are usually completely sequential and where sentences are related with each other, a large amount of text in the flyers is scattered and unrelated with each other.

Therefore, our model needs to take different parameters into account to extract the targeted data points accurately.

Some of the basic entities we aimed at teaching our model to detect and extract:

On a sample of 786 valid real estate data flyers, we have been able to accurately extract all the above-mentioned entities for 756 flyers which are resulting in a 96% success rate.

We explain this process more technically within our technical white paper hosted on Github currently.

Using different custom tools we can now import information into already existing spreadsheets, MLS, or CRM systems that brokers and others use.

Part 2 – Creating new possibilities

Automating real estate data extraction was a fundamental part of making a first step towards answering the following questions:

These are just some of the questions we are running further research on and we cannot figure it all out on our own, which is why we are inviting others to join us and use our research to build and discover further – because all our research is completely open-source

As we look to 2020 and envision the next decade, the traditional mantra of Location, Location, Location is becoming increasingly less relevant. The most successful CRE companies will likely be the ones that follow the mantra: location, experience, analytics. This requires companies to fundamentally rethink location, space requirements, users, and user preferences, and gradually shift to a service mindset.

Using analytics to create a better experience in process of buying, selling, and operating will need to be our focus in the following decade. Therefore, the speed of collecting and standardizing real estate data will be essential in opening a new chapter of real estate.

We believe there are large opportunities laying in a different kind of real estate – neural real estate. The team at Bonumic is eager to innovate further and share our research in order to eliminate current limitations of real estate discovery, estimation, acquisition, and speed so that we can, finally, focus on building a better-built world.

Subscribe for further updates on this and other research produced by the Bonumic R&D team.


Share to  

Facebook

Twitter

Linkedin

Get link