Opinion on Using Data Science to Conclude Real Estate Transactions


All Transform 2021 classes are now available upon request. Look.

The real estate industry is not the first industry that usually comes to mind when discussing how to apply. machine learning algorithms… The seller wants to sell the property and the buyer wants to buy it – it’s just a matter of closing the deal. The stumbling block is the negotiation of the price of this transaction. Accurate property valuation is a complex process that requires many different data sources and scalable pricing models. The buyer cannot simply refer to a detailed list of all possible factors and associated price values ​​and summarize all the characteristics of the property to calculate the total value.

An automated value model is a machine learning model that estimates the value of a property, usually by comparing that property with similar properties nearby that have recently been sold (“comps”). Opendoor real estate company relies on its version of the AVM – the Opendoor Valuation Model – to evaluate and find information about awards (for example, to understand the difference between the value of the award and the property in question). The company has invested heavily in data science almost since its history to enable various data sources and refine algorithms to improve model accuracy.

Speaking with VentureBeat, Opendoor’s Sam Stone explained why the company built the Opendoor pricing model and how data science fits into the real estate industry. With the company’s plans to expand from 30 to 42 markets by the end of the year and add new house types and price tiers, data science is expected to remain a major part. company strategy, according to Stone.

This interview has been edited for clarity.

VentureBeat: What was Opendoor’s problem and why did they decide investing in data science was the solution? What benefits did the company expect from scalable pricing models and data science investments?

Sam Stone: Since our inception, we have always been doing data science in-house and using both our own and third-party data for our models. We realized that modernizing the outdated manual home pricing process could benefit consumers in terms of price confidence and the ability to take advantage of capital in their home faster.

For most people, the house is their biggest financial asset, and they understand its value very well. It is very important that our algorithms include all the important functions of the house. Because every home is unique and market conditions are constantly changing, accurate home pricing requires constantly changing decisions. This means that we must invest heavily both in our algorithms and our team of in-house pricing experts to ensure that algorithms and experts run smoothly.

VentureBeat: What was already in Opendoor that made it possible to create the Opendoor pricing model instead of outsourcing work?

A rock: Accurate and efficient pricing systems are the backbone of our business model. Our original automated scoring model is based on lines of code that our co-founder and CTO Yang Wong wrote back in 2014.

Since then, we have invested heavily in technology and data side… We have developed various types of machine learning models, including acquiring and testing new datasets. We’ve developed processes for recruiting, growing, and retaining top-notch Machine Learning Engineers and Data Scientists. And at the same time, we have invested heavily in expanding our expertise, equipping our pricing experts with customizable tools to track local nuances in our markets.

It’s fair to say that pricing systems are at the core of our corporate DNA.

We always strive to learn from new datasets, new products, and new suppliers. But we have yet to see third-party vendors come close to matching the overall accuracy, coverage, or functionality of our internal set of pricing systems.

VentureBeat: Tell us a little about the Opendoor pricing model. What kind of data analysis and what investments were made to create this model?

A rock: The Opendoor Pricing Model, or “OVM,” is a core part of the pricing infrastructure that is used in many subsequent pricing applications. This includes our home offerings, how we value our portfolio and assess risks, and what decisions we will make when we resell a home.

One of the elements of OVM is based on a set of structural views of how buyers and sellers estimate prices and make bidding decisions for home purchases. They look at prices for comparable homes in the area that have recently been sold – often referred to as “compromises” – and adjust their home prices up or down based on how they think their home is equal. But how do you decide what makes one house “better or worse” over another? This is not a black and white equation, it is much more complicated. Homes have unique characteristics ranging from size and courtyard to number of bathrooms and bedrooms, layout, natural light and more.

OVM is supported by a variety of other data sources ranging from property tax information, market trends, and many home and area specific signals.

VentureBeat: What does OVM look like from the inside? What did you need to build to make it all work?

A rock: When we started building OVMs, we didn’t overcomplicate the task, relying mainly on linear statistical models. Starting with relatively simple models, we focused on developing a deep understanding of the thought processes of buyers and sellers. We could check and improve the quality of our data, instead of getting carried away with fancy math.

How we came to understand buyer behavior and salespeople have gotten better over the years, we have been able to move on to more complex models. OVM is now based on a neural network, in particular on an architecture called the Siamese network. We use this to build in the behavior of buyers and sellers, including trade-offs, adjustments, and weighting.

We have seen many times that the “modern” machine learning model is not enough. The model must understand how buyers and sellers actually behave in its architecture.

We have several teams of both engineers and data scientists who are constantly working on our OVM. These teams work closely with operators with deep field knowledge and often include them in product sprints. The development, quality control, and release process of our first neural network version of OVM was a collaborative team effort that took many months.

VentureBeat: What is the purpose of the human + machine learning feedback loop?

A rock: Our in-house pricing experts play a key role in making pricing decisions, working alongside our algorithms. We rely on pricing experts at different stages:

  • Adding or validating input. For example, assessing the quality of home appliances or finishing levels, which are important inputs, but difficult to quantify algorithmically. People do it much better.
  • Making intermediate decisions. For example, what features of the home might make it difficult to assess?
  • User-centric decision making. For example, given the set of homebuyer suggestions in our portfolio, which, if any, should we accept?

While we can more or less automate a specific area or task at a specific point in time, we have always believed that in the long run, the best strategy is to marry experts on pricing and algorithms. Algorithms help us better understand the strengths and weaknesses of peer reviews, and vice versa.

VentureBeat: What would you do differently if you were building OVM now using lessons learned from last time?

A rock: Ensuring high quality input data under all circumstances and for all fields is always the highest priority.

The model most accurate during macroeconomic stability is not necessarily the most accurate during an economic crisis, such as the 2007-08 financial crisis and the global COVID-19 pandemic. Sometimes it makes sense to invest in predictive functions that do not improve accuracy during “normal” times, but can be very helpful during rare but very uncertain times.

Last year taught us that we can evaluate homes using interior photos and videos provided by sellers. Before COVID-19, we personally inspected home interiors. However, when the pandemic began, we cut off personal communication for security reasons. As a result, we turned the interior appraisal into a virtual one and learned that it is actually much easier for sellers.


VentureBeat’s mission is to become a digital urban space where technology decision makers can gain insight into transformative technologies and transactions. Our site provides important information on technology and data processing strategies to help you run your organization. We invite you to become a member of our community to gain access to:

  • up-to-date information on topics of interest to you
  • our newsletters
  • gated content for influencers and preferential access to our valuable events such as Conversion 2021: Learn more
  • network features and more

Become a member

Source link