Learn how the most innovative company in real estate is leveraging Figure Eight to provide the best prices to home buyers and sellers
Opendoor prides themselves on their revolutionary approach to buying and selling real estate and their vision is to empower everyone with the freedom to move. The category leader and widely-recognized by media as the most innovative company in the field, Opendoor aims to simplify what’s often a stressful, grinding experience. Opendoor uses historical information on similar properties and customer data to power models that accurately price homes. They pride themselves on providing a fair market offer to sellers and spend a lot of time fine-tuning the models that output those prices. Opendoor helps remove the stress from what’s often life’s biggest transaction, giving sellers the confidence that they’re getting a data-based price and a world-class customer experience at the same time.
Opendoor has a big challenge: how do you value somebody’s home without seeing it yourself?
To start, Opendoor requests a lot of data from their sellers about upgrades and home amenities. For example, it’s not enough to know the size of a kitchen; they need to find out if there are granite countertops or stainless steel appliances or a built-in counter. Understanding these attributes helps Opendoor accurately value homes.
But data on home features isn’t enough. Opendoor also needs to understand the market. They need to know what a house in the same neighborhood sold for so they can teach their models what values in that area actually are. In many cases they get far more information from their own sellers than they get from historical transactions. Historical data usually has square footage and room counts, but doesn’t have information such as whether a house had granite countertops or stainless steel appliances. But historic pictures may be publicly available. And that’s where Figure Eight’s platform comes in.
Opendoor takes the images from the historical data on similar recently sold homes and runs them through Figure Eight’s human-in-the-loop platform. It works like this:
The historical data may simply indicate that a house has a kitchen with a certain square footage. But images from prior sales have details that are really important for Opendoor’s valuation model.
Figure Eight’s platform enables them to categorize these images, with labelers noting crucial items like the aforementioned stainless steel appliances or granite countertops.
By getting this data annotated, Opendoor knows far more about these historical sales than they otherwise would. They know a nearby house that sold five years ago had granite countertops and a clawfoot tub but not stainless steel appliances, for example. This historical data is a powerful source of ground truth for their valuation model, teaching it about what features affect home value in certain areas without ever surfacing Opendoor’s own seller data to Figure Eight at all.
Opendoor has begun using these Figure Eight annotations in Google Cloud AutoML and has seen some really promising results so far. After all, AutoML excels at image categorization tasks and Opendoor has hundreds of thousands of labeled images from Figure Eight to train their models.
Effectively, using computer vision instead of manual labeling allows Opendoor to leverage the data advantage they already have to automatically identify key home features that power their valuation models. This helps their models be increasingly accurate with every machine-generated label.
The combination of computer vision and high-quality training data will also help Opendoor move into new geographies much more quickly. Instead of annotating hundreds of thousands of data rows in a new locality, they hope to instead lean on AutoML’s vision models to provide a lot of those labels for them. And since those models were trained on accurate human annotations, the models themselves are already returning encouraging results.
When your business involves the biggest purchase most people will ever make, fair and accurate prices are of paramount importance. Opendoor is a great example of how to integrate different streams of available data and do something truly innovative with them.