Fed up with renting, I began looking for a home in an up-and-coming Oakland, California, neighborhood, which might have some semblance of affordability.
I searched for a home on Redfin. Though I have an upper price limit of $600,000 (this is California), I don’t use search filters. Instead, I bring Redfin’s map-based search to the neighborhood I’m looking for and click on a home and go to its property detail page.
It’s a Craftsman in my price range about the size I’m looking for. But the photos show that it’s a bit beat up, with ugly railings and old yellow paint. I want a more put-together home, so I continue my search.
Instead of going back to the map to click on a new home, I see Redfin has offered up a set of nearby homes just below the home’s description, high up on the page.
Predictive search (machine learning) at work. Wow.
One of the homes Redfin suggests looks more like what I’m looking for. I click on it. It’s nearby, has what looks to be fresh coat of stucco and beautiful tiled front steps. And it’s in my price range. I’ve found a home that I’d like to visit.
Trulia says only 6 percent of the millions of homebuyers who use its site to search for a home employ search filters, indicating that the online home search process is ripe for innovation.
Predictive search, made popular by Amazon and Netflix, is one way Trulia and rivals Redfin, Zillow and realtor.com are looking to improve on the real estate search process.
Their efforts helps may help some homebuyers find homes that they’re looking for faster — making agents more efficient and productive.
The big listing portals are all delivering on predictive search in slightly different ways.
Trulia and Redfin, for example, serve up two sets of suggested homes to consumers on the bottom of property detail pages:
- Homes the portals’ algorithms “think” users will like based on their individual browsing behavior, and
- Suggestions based upon the property attributes of the particular listing they’re currently viewing.
Lessons from ‘Trulia Suggests’
Trulia’s efforts in predictive search are an outgrowth of “Trulia Suggests,” an experimental approach tested by the portal starting in early 2013. Trulia Suggests prompted users to browse a broad selection of property types, and to “like” five homes that suited their tastes.
Trulia Suggests never came out of beta, Trulia’s head of consumer product, Jonathan McNulty, told Inman. In fact, Trulia shuttered the tool as a stand-alone feature this month, he said.
But the lessons learned from the platform will continually be baked into the portal’s search experience, McNulty said.
Trulia Suggests served as a laboratory to test ideas and make Trulia’s search experience better by serving up homes most relevant to users’ specific needs and desires, and connecting them with one of the portal’s agent advertisers, McNulty said.
The 18-month Trulia Suggests experiment has revealed some key insights.
While every user is unique, a home’s price, type and location are the core of every search. As might be expected, luxury items like a pool, a large backyard and other similar features are more expendable, McNulty said.
Not only is Trulia working on serving up more relevant results, it’s also tweaking how it presents them to users to spark the emotional connection that it has found is such a key part of the process of finding a home, McNulty said.
Role for Redfin agents
Redfin is also hunting for ways to make search a more intuitive and better process for the buyers who come to its site.
Redfin offers what the Seattle-based firm’s director of product, Andy Taylor, calls “collaborative filtering” — surfacing homes and neighborhoods “liked” by other users conducting similar searches.
The brokerage is also experimenting with ways to suggest homes that might be good fits for users who don’t employ well-defined searches or save their past searches, Taylor said.
Redfin’s perspective on search is a little different because, unlike Trulia, Zillow and realtor.com, it’s also a full-fledged brokerage, which means it has a particular focus on tying its own real estate agents into the search process.
Redfin has a machine-learning algorithm that collects listings that consumers might like based on how they search on Redfin, but it’s sent only to those who are currently working with a Redfin agent.
The tool, Listings Matchmaker, sends the collection of suggested listings to consumers’ Redfin agents first. The agents then give each property a “thumbs up” or “thumbs down” based on what they know about their clients’ specific preferences, and any inside knowledge they may have about a home that a computer algorithm may not.
Move’s foray into ‘machine learning’
Realtor.com operator Move Inc. is also on the hunt to revamp search.
Move is the sole investor in Ylopo, a search-focused startup co-founded by a Move exec that uses machine learning and data mining “to curate and recommend homes that we believe (users) will be most attracted to, thus creating a more intuitive user experience and a more interested prospect for real estate professionals,” according to the startup’s LinkedIn profile.
Realtor.com declined to discuss any details about how it plans to integrate Ylopo into its search technology, or how it’s experimenting with predictive search.
Zillow’s RealScout play
Zillow has also been largely mum on how it’s thinking about predictive search. Currently, it serves up a set of three “similar homes for sale” on the right rail on property detail pages, but didn’t share how those properties were chosen.
RealScout, a startup that aims to level the real estate search playing field for brokers and agents by selling them access to its search technology, is also squarely focused on matching homebuyers to homes.
By building elaborate tags for listing images, spotlighting the proximity of neighborhood-specific amenities and constructing detailed homebuyer questionnaires, RealScout, which announced a $6 million funding round in November, is looking to serve up better search results, tailored to specific homebuyers.
Predictive search’s mobile gateway
Mobile devices, which can trigger instant notifications based on their locations, are also a big part of the coming tide of predictive search in real estate.
In May, for example, Trulia updated its smartphone apps to alert users to new listings that meet their search criteria when they come within a certain distance of them. The feature’s geolocation element is a big part of Trulia’s effort to move users from search to discovery, McNulty said.
In February 2013, Zillow began powering the real estate portion of Google’s intelligent personal assistant, Google Now. When users enter keyword search terms that indicate they are home shopping, or make frequent visits to real estate sites, a new Zillow “card” for Google Now provides information on nearby homes for sale and open houses.
Redfin’s mobile tools allow agents to quickly filter through and assess the customized suggestions Redfin sends to its buyer clients, Taylor said.