Ai & ML to Transform buying and selling a home
How buying and selling a home could soon
be as simple as trading stocks
Artificial intelligence in housing could
completely change the way we buy, sell and live
iBuying was just the beginning. Get ready for
machine learning to remake real estate altogether.
By ANDREARIQUIER Published: Sept 11, 2019 10:30 a.m. ET
On a recent weeknight, Dahlia and Adam Brown came home to a
spacious Colonial on a quiet cul-de-sac in Marietta, Georgia. The Browns both
work demanding jobs and have two young sons. They bought the house in June
using Knock, a company that’s trying to revolutionize the real estate industry
with a “home trade-in platform” making it easier to buy and sell at once. That
solution was ideal for the Browns, who are just as busy as most couples, but
are more introverted, making the idea of prospective buyers tramping through
their private space seem excruciating.
Across town, Martha Seay was overseeing movers in a rambling
brown ranch-style house nestled among tall hickory trees. The day before, she
had closed on the sale of the house, where she and her husband had raised their
family, to Zillow, the massive real estate company. The next day she would
leave for Florida’s Gulf Coast, where they had just bought a retirement home.
Seay had wanted to move for years, but the idea of selling was
daunting: “I said, maybe next year, maybe next year, maybe next year, because I
didn’t want to go through all the crap you have to go through.” Selling to a
company took just a few clicks and one visit from an appraiser, and Seay was
delighted. “I cannot tell you how much the stress was relieved,” she said.
The Browns and Seay are the consumer faces of the disruption
that’s currently roiling residential real estate. As different models — home
trade-in companies, “iBuyers,” partnerships between new upstarts and old stalwarts —
clamor for attention, lots of attention is focused on trying to determine
what’s here to stay and what’s just an awkward rough draft: the Pets.com of the
housing market.
But these families are also part of a massive industrial revolution.
Information technology has remade old processes as different as ordering dinner
delivery, hailing a cab and trading stocks. Now it’s coming for an industry so
last-century that much of the paperwork is still done on paper, where customers
are often steered among professionals scratching each other on the backs, and
where there’s enormous incentive for incumbents to keep making it hard for
customers to do it themselves.
The stakes are big: $74 billion of real-estate agent commissions
were paid out in 2018, and investors have poured billions into all kinds of disruptors. Early
adopters like the Browns and Seay give us a glimpse of what the future real
estate market could look like. But just as online retail has hurt the
bricks-and-mortar retail industry, and tech-enabled social networks have
changed not just high school reunions but may be influencing the political
process, data-fed real estate could upend our lives in many ways, some we can’t
even comprehend yet.
“There’s over 100 million active users on Zillow and Trulia every month but only six million
people buy and sell houses every year,” said Charles Folsom, Knock’s director
of customer service. “Even if they’re just window shopping, there’s clearly a
desire there. If you can empower the American Dream and enable mobility at the
same time, that’s the best of both worlds.”
Zillow, a company that keeps an uneasy truce with real-estate agents even
as it increasingly tries to automate the work they’ve done for decades, may
have even bigger ambitions.
Krishna Rao, Zillow’s head of analytics for its Offers division,
likens the current evolution in real estate to the democratization of stock
trading a few decades ago. Not only is it possible to look up the value of any
stock instantly today, he noted, but “there’s a kind of perpetual bid-ask
spread on every stock, right? I think we’re a long way away from that in the
real estate space, but how do we take incremental steps toward it?”
Zillow sees the listing price
as a ‘machine learning’ exercise
In 2018, Zillow took what had been a small pilot program and
announced it was going whole hog into iBuying, the practice of buying homes
directly from consumers. (The term iBuying is also sometimes called “instant
offers”; Zillow’s program is Zillow Offers.)
Rao, a macroeconomist by training, had joined the company in
2013 after a stint at the Federal Reserve Bank of New York in the thick of the
financial crisis. At Zillow, he helped analyze and make useful the enormous
quantities of data the company captures for the Zestimate and other kinds of forecasts and
reports.
In the second quarter of 2019, Zillow bought more than 1,500
homes and sold nearly 800, and says it aims to transact 10 times that amount. Rao’s group is
in charge of thinking about how it should all work: What should the company pay
for a home? How much will it sell for — and what should it be listed for? How
quickly will it sell? What upgrades are necessary and which contractors should
be dispatched to do the work?
More to the point, when your “inventory” is dozens of houses
scattered around a sprawling metro area, with the constant threat of mold,
floods, power outages, unmown lawns, downed tree limbs, etc., who’s keeping an
eye on the goods? (Rao told MarketWatch that Zillow is currently recruiting
high-level logistics people from the likes of Amazon and “classic industrial
companies” like General Electric to make this transition.)
The promises — and the peril — of this new endeavor are weighty.
Zillow’s stock tanked after its last earnings release,
in which management revealed that a small sliver of the homes it had purchased
were being held longer than they had accounted for.
Analyst Brad Safalow, who has a short position on Zillow shares,
betting on a decline, wrote: “Even a 10% hit to the company’s inventory could
cut Zillow’s overall gross profits from its Homes division by 25%! The margin
for error in this business is razor thin, and we think investors continue to
underestimate the difficulty of this ambitious endeavor.”
But Zillow bulls, and management, point to what Rao calls its
“competitive advantage.”
Lots of companies have housing-market data about supply — that
is, listings of homes for sale. Zillow’s secret sauce is information about
demand, gleaned from 180 million unique website visitors each month. “That is,
seeing who’s searching in this neighborhood and
are they also searching in that other neighborhood or are they really just
pinned down in this area. What is the demand for three bedrooms like relative
to four bedrooms?” and so on, Rao said.
What does that mean in real life? Zillow sees the listing price
as a “machine learning” exercise, he said.
“That machine can look at what the relative demand is for homes
like this, relative supply, how that’s trended, and take these gobs of data and
crunch it down into a particular listing price. Over time, as that home is
listed, we then get more and more granular information — how well is the home
showing? Are we seeing lots of tours, lots of offers? And use that to refine
our strategy.”
“How do we solve the problem of
consumers’ pain”
In a shared office in Buckhead, a well-heeled Atlanta suburb,
the Knock team is working on the same questions. Two of Knock’s co-founders
also started Trulia, a Zillow competitor that the bigger company eventually
bought. Both companies launched as the housing bubble was peaking. Zillow
quickly became known for the “Zestimate,” a modern marvel of housing clickbait
that made the value of a home, previously something an owner considered only
infrequently, a near-real-time interactive experience. (The Zestimate preceded
Zillow’s listings, while Trulia started by offering online listings and later
developed its own home value estimate tool.)
“At Trulia they unlocked the database of listings and now
they’re unlocking the other side — how do we actually solve the problem of the
transaction,” said Stephen Freudenberg, Knock’s first employee and a former
real-estate agent. “Most of these other companies are solving for the agent’s
pain, not the consumers’ pain.”
Knock does that by helping customers buy a new home — usually a
larger one to accommodate a growing family — then sells the old one once
they’re settled, and out of the property that needs to be staged and shown.
They charge a fee equivalent to 3% of the value of the home they helped their
clients buy, and 3% of the cost of the house that gets sold, as well as a small
surcharge to cover the costs they front their buyer clients, such as initial
insurance and escrow payments.
It’s a personalized model, almost like a concierge service. Yet
Knock seems to spend nearly as much time and energy on data analytics,
specifically about price, as Zillow does. The company recruited its lead data
scientist, Rafaan Anvari, from the Central Intelligence Agency.
Anvari spent months shadowing Freudenberg, asking a constant
stream of questions about how and why Realtors do what they do to create an
automated valuation model for homes that understands even better than a
seasoned real estate agent how to gauge pluses, like access to a golf course,
against minuses, like proximity to a busy road.
Their back-and-forth went on for months, and some of the
futility of getting a machine to learn how to think like a veteran neighborhood
salesperson are captured in their internal chats, as seen below.
Their
automated valuation model is now named “Borg,” after the drone-like cybernetic
beings that tried to “assimilate” humanity on “Star Trek.”
The Knock team doesn’t just think Borg will make them more
competitive. They think it will solve a lot of what’s wrong in today’s housing
market.
“Ask five different agents what your house is worth and you’ll
get five completely different answers,” Freudenberg said.
Internally, Knock team members call the existing real-estate
ecosystem a “gypsy market” because it’s so antiquated and opaque.
“Everyone’s haggling but they don’t know what they’re haggling over,”
Freudenberg said. “They’re just making up obscure numbers.”
He offers an example: A family might spend $100,000 remodeling a
kitchen but add only $50,000 to the cost of the listing because properties in
the surrounding area, which are comparable listings, might not have such fancy
kitchens. “So they’re stuck with what the neighborhood sold for, but if we’re
actually looking at the data then everyone could theoretically get a better
deal.”
Borg plugs information including room sizes, home style, outdoor
space and more into an algorithm to derive a home’s value. Meanwhile, Zillow is
trying to get even more granular, by teaching its machines about internal
fixtures and features. The company described that evolution in a July press
release about the Zestimate: “The image-recognition model can classify patterns
in the pixels of photographs and correlate them to home value. For example,
while the human eye sees tile or granite countertops, the Zestimate identifies
two different pixel patterns.”
It’s worth noting that the vast majority of data-science
resources in real estate seem to be focusing on home valuation as the endgame,
as least for now.
Rao suggests that may be “because it’s a very narrow,
well-defined problem, so it’s kind of easy to show progress to investors. We
think of the strategy of Zillow Offers not just as a crisper valuation, but
kind of an end-to-end experience that can seamlessly integrate the mortgage
piece of it, the title, the escrow, and the buying and selling. It’s a big
challenge doing all those things at the same time.”
Still, revolution has to start somewhere. The industry’s focus
on automating valuations means that very soon, the Federal Reserve is likely to
finalize a regulation that says appraisals will no longer be
required on most property sales up to $400,000.
For Dahlia Brown, the Knock customer in Marietta, having an
algorithm at the heart of the real estate market may help counter some human
bias by limiting “some of the historical practices that maybe have kept certain
people from homeownership,” she said. “This process actually seems as fair and
equitable as it could be.”
Jeremy Sicklick is CEO of HouseCanary, a platform that’s raised
over $60 million from investors to help it aggregate what he calls “millions of
data elements” to come up with accurate home price valuations, forecasts of where
those prices are going, and rental valuations for those properties.
With HouseCanary, investors, like those who buy single-family
homes to rent out, can see a property status change, like a price change or a
default, in near real-time, Sicklick said. “We’re helping them identify real
estate opportunities within five minutes,” he said. “We’re getting into a world
of programmatic trading in real estate with large institutional investors that
we help enable.”
With so many concerns about institutional investors like Blackstone
snatching up homes to rent out, keeping low- and moderate-income Americans locked
out of homeownership, is that a good thing?
“I actually think it’s a great thing,” Sicklick said. “What I
think it does is it adds liquidity into the market. The iBuyers and the large
institutional investors create a very liquid floor price for someone looking to
sell and drive a faster transaction.”
There has to be a better way
For now, that model is as imminent as, say, driverless cars.
Knock, Zillow and other upstarts are still plying their trade in a mostly
20th-century housing market. The Brown family, for example, contacted Knock,
got pre-approved for a mortgage, had their house assessed, and started touring
new homes within a week. But they hadn’t even moved out when their agent urged
them to hurry to get the property prepped for sale, to take advantage of the
waning spring selling season.
The Browns are happy with the outcome and say that having the
process rush by so quickly — start to finish in two months — lessened the
stress of having to do precisely what they dreaded in the first place: let
strangers poke around their home. But it’s still worth noting that traditional
cycles of demand and habits are trumping the potential that new models offer,
at least for now.
The question remains: how much of what comes next comes down to
algorithms, and how much to process? For now, the onslaught of machine learning
in the housing market continues unabated. As Knock’s data team might say,
resistance is futile.
Comments
Post a Comment