Lessons To Learn from Zillow’s AI Bust
By Digital Aptech - Nov 23,2021
By the end of October, it was officially news that Zillow had shuttered what seemed to be a promising enterprise: Use an algorithm to determine what houses the company should buy with the intent of flipping it for more money shortly after. Zillow Offers was its name, and it was laboured by over 2,000 people and ended up representing a loss of $500,000,000. The company assumed that it had a treasure trove of real estate data on hand, and a very promising love affair with new-fangled AI technology, would result in untold profits. Unfortunately, through a mixture of poor business decisions and the limitations of AI in the face of random events, the company instead had its market cap move from $48.35 billion to $16 billion.
Understanding Zillow’s Approach
Zillow invested heavily– both in capital and manpower, into the development of an outfit of the company whose sole focus was flipping real estate based on data coming in from an algorithm. This house flipping would usually involve making some small renovations to the homes that get bought up, with the intention of re-selling them in a very short window rather than holding on to them to accrue value.
They assumed the money they’d make would be from things like transaction fees and services the company offered, but found it was making a fair bit of money from flipping as well. This completely switched around by the summer of 2021, however. Zillow was following the advice of its algorithm and buying a plethora of homes at a relatively high price, right around the time that price increases were starting to slow down.
For example, real estate price increases in Ontario have experienced a minor slow-done in October. See our overview on Condos that are set to be occupied come 2022 for an idea of currently estimated prices.
Other Companies Trying Algorithms
There are still other companies that haven’t lost faith in the ability of algorithms and AI to guide investment decisions, and as a prospective investor you might want to keep tabs on what these companies are up to. These companies are following a trend of trying to minimize the human element in the buying and selling of homes. The idea is that a person can use their service to put their home up on market and have it bought in a short amount of time without needing to hire a middle-man realtor.
This technology piggybacks off the rising concerns of real estate agents playing unfairly with their clients, with examples in recent news of foul play. It’s quite possible that some companies will manage to avoid Zillow’s failure. It’s important to know that while this blog argues for the value in human realtors, the technology for non-human services that facilitate real estate sale is improving and will be disruptive.
Until such a time, a lot of the heavy lifting continues to be done by realtors, developers, and the sites each refer to for clients and information.
Why Humans Matter
For a lot of other sectors that deal in large flows of money, like handling foreign currency exchange or the stock market, smart algorithms are becoming king. It’s easy to want to adopt this to a field like real estate investment because of how it can be similar in some ways– companies’ stocks are affected by market trends. In Ontario, the cost of a new home in Hamilton will be directly impacted by things like how much immigration is coming to Canada, how popular is Hamilton as a landing point for that immigration, and even how much wood should cost next year vs. today.
In a perfect world, these are factors that a super-intelligent AI can gather together and spit out real, actionable results with. For example, there is a new-ish field of AI design tools that generate architectural blueprints for buildings, or models for instruments and machinery, based on a number of factors you can put in. But compared to factors like “building must occupy this square footage, be this tall, link to this plumbing system”, things like the global economy can be pretty dramatically altered by seemingly random events.
Where The Algorithm Failed
There were a lot of issues in Zillow’s approach that demonstrated the value in having a human touch.
An algorithm can understand that a home has 4 bedrooms, but isn’t able to measure the actual layout of those rooms. How big are they? How high is the roof? Do they actually look any good? Are there any concerns after inspection of mold buildup– is the house teetering over the edge of needing to be practically rebuilt? A lot of the time, a home needs a human being to be standing in it’s interior, walking around, and getting a personal measure on how well laid out the place is.
The Algorithm was meant to be able to take advantage of the huge pool of data Zillow already had on how well homes had sold in different areas, for how much, and what sort of price changes happened over different periods.
Part of it was human error as well. Zillow was actively paying more money than what the algorithm might’ve recommended they pay, in an effort to catch up to other companies who’d invested heavily in the flipping game. News reports came up in which a house was bought for well over listing price, and so on.
Dealing with Randomness
Generally, humans are pattern-recognizing animals. At the same time, they’ll misconstrue things and think they see patterns that aren’t really there.
Real estate agents are people who’ve specialized in looking out for patterns of things that might affect the price of real estate. For people who aren’t officially licensed agents but still want to get into the game, some light studying on the subject can get you quite far thanks to the number of free resources available on the net. Looking into historical data, taking a look at real-time conversation happening on public forums like twitter (as long as you can vet the speakers as actually knowing what they’re talking about) and following good news sources are really important.
A certain blog post might talk about fantastic opportunities for new homes coming up, and you can follow it up by googling around about the town the blog is talking about. It’s good to pick out sites you feel you can rely on to provide accurate information on real estate info on a city by city basis. If you’ve heard in a news article about a town like Prince Edward County, you could check out our page detailing the real estate landscape for new homes in Prince Edward County. Condos HQ in this example will consolidate all sorts of useful information from different sources to paint a useful picture of what it’s like there.