Modernizing Real Estate with Data Science // Ian Wong, Opendoor (FirstMark's Data Driven)

[Music] thanks man for having me and it's really good in New York it's you guys have long very heavy rain here hand you know all of us here have gathered here because we care right because we are data scientist we are entrepreneurs we're problem solvers at large I want to work on problems are both interesting and hopefully that makes a big impact in society and an open door we build a company around a really hard problem that hopefully is very meaningful and impactful and that problem is how we simplify the way that people buy and sell houses how do we modernize the real estate experience and how do we do that with data science now a lot of you are looking at me like what's in talking though why do we need to modernize the real estate experience so I know I'm talking to a lot of New Yorkers here so settle your shoes for a moment and pretend that you are a typical homeowner in the country okay so I'll just pick Phoenix for example so you're that home owner and your home is worth about $230,000 and it represents the majority of your net worth now you have to move okay so maybe you're expecting a family so you're sizing up maybe your kids are moving away you're sizing down maybe you're going through a divorce maybe a shop relocation whatever the case might be you need to unlock the capital that you have in your home and so you better start on day zero and you decide you know what I got to I got to do that well guess what it's going to get over three months and that's in that three months period you don't actually know when you get too close and you don't actually know how much you would close for on top of that you're going through open house after open house you're negotiating and there's still a 14 percent chance that a deal might fall through that is an incredible amount of stress for the largest transaction of your life and so what we're doing here at open doors we want to simplify that and make it very easy for people to himself and the way we do that which will actually make an offer for a house and buy it from you so you don't have to endure three months of stress and headache more on that in a sec so the average loan takes over three months of sell and there's a one in seven chance to deal with loss through and our goal is to enable homeowners to sell instantly and eliminate these three months of stress hassle risk and uncertainty how big is this problem well it turns out it is the largest asset class in the United States roughly two-thirds of Americans are homeowners it's a single biggest thing that we spent money on every year one point four trillion dollars of assets change hands generating a hundred billion dollars in fees this is a massive industry in which consumers aren't really left with that many choices right and that's what we're here to do at open door we want to really create a new option for sellers and so here's our clock if you're a homeowner in Phoenix Dallas or Vegas our first three markets you can simply log on to a website open or calm submit your address sell a home profile and within the same day we'll give you your offer and here it is will tell you hey this is the fair market value for your house and here's the amount of fees that we're going to charge and we're being very explicit with that and you know exactly how much you can take home and if you click Next we can close and as quickly as three days or so what we've done is eliminated three months of stress and hassle and just oh my god what I'm going to do with my life and you can basically move on whenever you want if so that's the product now in order to deliver on this product premise we have to sell some really hard problems cheap which is how do we price our most valuable assets how do we create a pricing index that clarifies the value of these 25 trillion dollars worth of assets you know you and I can log on to our bank account and check our balance down to the cents but when it comes to the biggest thing that we own certainly we have no idea what it's worth so how do we flip that equation a little bit and so here's a map of Phoenix Parkins and here are all the transactions that happen in Phoenix in the last year and the sizes of each dot is scaled relative to the amount of the transaction so the name of the game for us is we have to understand and estimate perfectly the size of these thoughts until II might ask a in like don't people do this already today why do you need data science well yes today if you're a homeowner you can get a broker price opinion you can get an appraisal and you will pay through three to five hundred dollars for an appraisal for instance and that will give you one opinion of value for your house for this moment in time in order for us to deliver on our product premise of simplicity certainty and speed we need to be valuing all these houses at any given moment in time right and not just in this instance we actually have to estimate backwards in time and actually forecast forms in time because we actually buy these houses and we have to manage inventory risk right so we can so hopefully if people here can appreciate that this is on a whole different order of magnitude when it comes to difficulty and that's where data science comes in and so what are some of the challenges with the designs and now I want to talk through highlights three of them and as data science it's a general predictive the program to build predictive models is that we play pretend we pretend that we are a buyer in this market and we ask ourselves where things that we think about and our job as a scientist is to take those intuitions teach our model those intuitions and see which one of those generalized right which one of those would stand the backtest let me highlight three problems that you encounter if you were to ask yourself these questions so the first one is you want to see the home second one is you want to study the neighborhood a third ones that we need to look across time so if I ever tell you hey there's a home in Phoenix is 2,000 square feet for that two-bath can you give me a value like no I need to see the home ok so that's the challenge a we knew start by actually being returned very visual signals into something that not is machine readable that we can show and ask the algorithms to learn from and with that forward that - bass 2,000 square feet home that was built in 1995 as a home like this now suddenly you're getting all these sensory signals right things that are as they relate to the quality of home that matters when it comes to the price of the home so what arts only some of these features are felt fairly quantitative and others are actually quite qualitative so the quantitative ones is there a pool in the backyard what's the type of the flooring what's the cabinetry and stainless steel appliances these are actually fairly quantifiable and then there are other ones they're actually much harder do you look do you like the look and feel this house what is the curb appeal for this house right so our job and data science is to then turn all these visual signals into something that's very structured like how do you quantify curb appeal that's a pretty difficult challenge and so practically speaking what we do is combination of deep learning so we have a couple boxes and Amazon that's running cares and it's kind of turning all these photos into structured data and also a lot of crowdsourcing I know that deep learnings in both of these days and everyone's talking with people Orion provides Oklahoma people are in talking tonight um but I'm really hoping that Justin 17 at the site know it's going to be the come back here for crowdsourcing because we need it and it's totally unappreciated of the technique okay great so we can turn these visual signals into structured data are we done can you now price this home if I tell you that you have these photos well no because you want to actually study the neighborhood - right they say in real estate location location location okay so let's take a look at where this home is located and here's the math of a neighborhood in Phoenix and in the back of your mind you're like well you know cookie cutter home and suburbia Phoenix what is so difficult about pricing it's just suppose you know we're trying to price this home that's got a white arrow here and these other homes are one to transact in last twelve months right and you're like you know what I can probably triangulate based on these other homes what they close for what this home in the middle should should be priced at but really in reality housing is actually quite heterogeneous so there are a lot of features here that are quite variable that would impact the relative prices of these houses let me point you to a few of these so on top here we have a home that actually backs into the sixth plane highway and that will obviously be inconvenient and noisy you can have a higher line situation over there so that's probably not a great look in your backyard others are talking on coldest X was probably a bit quieter some of that pools from F not your front yard some have football fields in the back you can imagine that this problem gets complicated very very quickly right and this is not the traditional big data problem and the traditional big data problem you're you're you have basically a lot of data right so a lot of n really in this problem we have very wide data are join key in our data so it's literally this parcel right here right and you can create any kind of data or any kind of features that derive from this home all right and that's the difficulty about this problem it's a very very wide data problem on top of that there are real sparse to the issues in this problem as law so in any given 12-month period there aren't actually that many transactions that happen say in this neighborhood so if I were just to take a snapshot and look at a block group and look at a number of transactions that happen there are only a handful so what that means is you actually have to use sound inference techniques to understand the distribution of values for any given home right and now we're starting to walk in the rarefied territory of modeling and most people haven't really gone to this phase in terms of thinking about how to download their houses and this is just for spot price right this is just what I think the home should sell for today and that's just level one level two is how do we done actually telescoped is both backwards in time and forward in time but we need to now start thinking about how we can do a little bit a little bit of time travel so here I'm going to share I want to contextualize the challenge here and I want to share data vids that shows the last 15 years of transactions in Phoenix so here it is Rhett means transactions that close below the historical average Blue means transactions close above historical average and in the early 2000s you see a pretty good mix of red and blue so it's pretty even-keeled until you see the bubble years when certainly all the transactions start reaching historical highs as they all flash blue and you guys are all holding your breath here and the subprime crisis hit and then certainly a problem telling prices started coming down drastically until around 2012 when the prices start to recover and we're now on a more stable long-term trajectory so my point here is that the same home that same thousand square feet home depending on when you ask the question what that value is well that you have a very different answer if you ask that question tested for 2008 2012 or today right so there's a lot of heterogeneity not just in space but also in time and we have to model for that into what are some of the challenges modeling here is the same view of that data set now with time compressed in one dimension and you have to close prices and each dot here to transaction that closed and so what you see here is for any given moment in time to pick a time splice here along the you know along the line here you see that there is a struck there are structural differences in homes that make it close for less or more maybe sometimes for smaller some ones are bigger so you have patterns along this axis and you have clear patterns on the x axis as well you have trends both cycler and seasonal and they're even look really tiny small little kind of factoids if you all things that you've to work with which is that people like to close and round numbers so you see these striations and is because people like to close for 200k or 201k when in fact enough that's just totally just human emotions that way so those are criminal challenges another civil challenge is that we're dealing sparsely again so you can just look at the closed prices and you average that forward zip code and attend 28 day time frame you see this really kind of volatile blue curve and again we have to think about how to really wrangle with sparsity here and one last time series problem I'll mention is that not only do have to think about price we also have to think about liquidity it's not just a question about volatility and in where the prices would be you actually have a model what is the market absorption rate and how ready is the market willing to accept a home that we're able to buy and resell I'm going to price that and that's part of that risk there are a lot of really really hard problems that I've pointed to and as a start-up you know we've we have been successful so far but really you know we're only we're not even three years old yet and we you know have a team of 220 people 40 and technical team and roughly half of them learn data science so not only do we have to solve the data science problems themselves we actually have to solve a meta problem of how do we actually build the team so there are a lot of things that we are a little bit different in terms of what we look for in our team and how we execute and I know I'm kind of running out of time here but in the in a Q&A period I can give you my rent on Venn diagrams Matt and we can talk more about that later so these are some of the things that we look at when we think about recruiting for our team and when we think about running our team effectively and really you know we do we solve all these heart problems so that we can deliver on our product promise the thing that gives me frankly the most satisfaction when we build a team so all these hard problems is that we get to help our consumers move on to the next stage of their lives right so so far we have helped thousands of consumers so far to to kind of sell their home and lock that equity and move on and we look forward to doing that for many others and with that I'll conclude thank you very much fascinating actually would be very interested if you wanted to comment on that team slide the Venn diagram I like the type of people that are good for this yeah it's interesting because there's these these Venn diagrams I think people taking too far and you're the ones that have the I'm a hacker and a computer scientist and a data science or sistex and then that sweet spot in the middle I'm a data scientist right I think people take that a little too far I think that's the minimum bar to be a sufficient data scientist well a B it totally under plays importance to software engineering and C that's that's actually just a small part of what it means to be a scientist I think I'm totally neglected things like business acumen the ability to actually just solve a problem you're and what people do is they bound themselves in these boxes like I'm a data scientist I will only ship models and api's know like your data scientist your job is to solve really hard problems all right and so people box themselves in and let talk to young especially young gay scientists they're like I've got all the skills I need and it's done diagram like oh yeah it was a Venn diagram you're getting boxed in by the Venn diagram so that's my quick rant on that very good thanks we woke Appliance who have similar assets on the financial sector but can you comment on when you're the avian technology that you have developed when it gets a bit more commoditized five years how would that impact this asset class on the institutional side yeah I think there's the there's an implicit assumption that the AVM will become commoditized right and I think and regardless of whether or not people come up with various value opinions what really matters is that we are able to actually turn back I think what really distinguishes open door from the rest of the companies out there is that we actually put our money at work right but we stand behind the model and we do that by actually buying the asset from the seller and then reselling it so you know what really matters is whether or not we are accurate and we can deliver that for the mass volume can you scrub the process to resell afterwards and the team that you have involved in that part yeah so what do we do after we buy all these tells us we just you know hold on to and forever yeah so what's really interesting about open doors once we buy all these houses it really you know this inventory while it's you know obviously with the managed at risk it really allows us think about how do we optimize the buying experience because today which I talked a lot about the selling experience but the buying experience is equally hard right so things like accessing the home finding out what the right house for you and then I am I going to get up get home these are all pain points so we have on the resale side we have a team of both operators engineers and design isn't what are they doing they're trying to find the recommended the best homes that fit on the buyer they're building system like like all the open houses a lot of people text to enter any of our home st. end they want we're building programs like trade in so in the same way with a car you can trading a car why can't you do that with house we built that on top of that if you think about making this really big purchase how do I know that I want to get a good home right you're really banking the fact that that inspector will tell you all things you need to know so what else we even have something to call a 30 day satisfaction guarantee so you can actually return to home to us if you don't like it on top of that what's really interesting is because we are acquiring at this point hundreds of houses a month what we can do is we can we have the best inspectors the best G fees the best specialists who are servicing the houses going transfer those quality assurance to the sellers or to the buyers rather and so we have we provide them before in T as well so we're doing a lot of work to innovate on the buyer side as well so how are you hedging your financial risk yeah well I guess it's questioned a lot which is can you buy some kind of hedging instrument and really it's there are not that many available out there and they're not liquid enough where they're attractive from a you know to actually turn back on and so we what we have to do is actually be very accurate now the offer that a seller gets has two numbers in them one is the fair market value so this is what we think your home would self on the market and the other one it's a fee and that fee actually flexes depending on the risk of the home and so what we have to do is actually the price works very well and so the subtext on the title slide which is how do you probably price the most valuable asset actually the price both in terms of the fair market value and also but also have a really good grasp on the risk profile hi terrific presentation you have a very interest in the financial risk side as well in the selling side do you try various things to optimize what you do for example you could potentially rent houses for a while too if you think that the price is going to go up in a different season you do you try to furnish the homes when you sell them to improve the resale value most people know that if an empty home doesn't usually command the same price as a furnished one I'm just curious what you do to kind of test and optimize the selling process for your your homes we do a lot well and think of us and the resale inside is very very sophisticated seller so we have a lot of inventory and we'll try all sorts different things should we stage should we not how much attention should be given a first day of open houses how often should we refresh our listings these are all things that we're constantly experimenting with to optimize for the resale do you take into account the floor plans at all like coming from New York City where you have bathrooms that are is 30 inches wide and like bedrooms that barely fit it will go bad so you take it four times yeah for sure I mean that's what's really interesting about this day a science problem right because there's a multitude of things I could impact the value and what we have to do is to turn all these things for plan include it into structured data so yeah we definitely take a look at floor plan and it's really interesting and a part of the point is they find just working with operators working with folks that who do this day in day out and what are you looking for in the floor plan right how do I actually encode that in in the model but yeah we just wanted to go over that so at the end of the day if I have a house to sell why should I trust that your price is the right one it's a great question we get that a lot over time and all this is public data right so the recorder office will record the sale and in the long run as we have cumulate more and more samples people will see our track record right and so we're playing a repeated game here we're not trying to play a single game where we're trying to optimize the margin for this one instance we're trying to bulb the reputation of becoming the fair market value the provider of fair market value and so the short answer is well a short answer is that this is all in the public domain and you can actually check to see that we're accurate hi the first one first things you said is the tenure not in New York and you're in Phoenix we're obviously all new your motives from New York there have been certain companies as specialized and done well in New York real estate because there's some very very specific issues that affect New York real estate have you done in New York at and these are problems that also have similar problems in other specific locales like New York for example proximity to school zones or it's a very complex issue proximity to subways somebody else mentions bathrooms have you done in New York have you considered some problems in other cities yeah it's a good question we consider New York we consider Semak let's go I think eventually for sure I think the thing to appreciate about this industry that is that it is really really big and there are many many cities outside of SF and New York and the average I mean it's crazy cuz we all found out recently is too soon too soon oh okay but um the average price of a home in the country of $230,000 right but when you list a home in the color anywhere else in the country you actually close for three percent less than you listed for right like we have actually designed for this case that up that works for majority the country before we enter kind of SF in New York because if you optimized for SF in New York you might actually be able to enter the rest of the country alright one last question and then Lydia was there thank you actually this subject is really interesting to me because I'm looking to purchase a home in Ann Hatton or on DC's in like the most prices areas as a first-time home buyer but my quick question is like let's say someone utilizes your services and they sell their home to you what if that seller down the line hears about his neighbor or multiple neighbors selling their comparable home at a much higher price than you off you offered and provided to them do they have recourse or not I'm just wondering like down the line but it's an interesting question because what we're trying to do is press fair market value so if over the course of six twelve months if the market appreciated is on us to provide the fair market value at that point and it's kind of like you know if you were to sell your home and you have a buyer that buys it for a million dollars and three months from now the market appreciated and your neighbors bought their home for 1.2 will you get mad right that's the same dynamic at play all right very good so since we're at the end of the event and people can come and ask you questions directly we'll be do this because it's a terrific thank you very much [Applause]

Channel: Data Driven NYC

Views: 12,065

Rating: 4.8720002 out of 5

Keywords: startups, technology, big data, data science, venture capital, real estate, machine learning, data, opendoor, entrepreneurship, entrepreneur, analytics

Id: dR5N8cMkIGQ

Length: 24min 25sec (1465 seconds)

Published: Fri Feb 03 2017