For quite some time Zillow.com has served as a one-stop site to manage a real estate buy, sell, or rent transaction. By visiting the website, anyone can find all manner of properties for sale or rent. They even maintain an inventory of residences in dozens of cities throughout the country for the purposes of making instant offers to prospective buyers. The entire activity was controlled by an automated process of algorithms that enabled the company to flip houses for a profit. Then suddenly in November, 2021, they shut the instant-buying machine down. As it was still a good time to sell houses as it had been over the last two years with prices rising, why would they close down that seemingly lucrative business activity?
The process of i-buying real estate: There are different models of the quick in-and-out trading of houses, but they all thrive on two basic principles: the current price of the property, and a forecast of the price a few months down the road when it is expected to turn over. To maintain this information flow, a large body of data is required, and needs to be constantly updated. Such data elements as the exact location of the house, number of rooms, pool yes or no, square footage, etc., get compared with the same data of houses that have sold in the area recently. The output helps the market move more smoothly by limiting transaction costs. It also facilitates the buyer with both ends of his transaction—selling his existing home and buying a new one in a different location. The algorithmic based forecasting enabled Zillow to make instant offers to the buyer for either or both ends of the transaction.
What went wrong: Zillow’s management admits that the problem lay in its inability to accurately forecast the value of properties three to six months down the road. Generally speaking, for forecasting to be meaningful it requires certain assumptions to made and to come to fruition. As market conditions shifted, the forecasts were unable to keep up, and Zillow found itself trading houses for an average of 6% below their cost in some areas. To make matters worse, the firm had been loading up its inventory in a frenzy of house buying over the previous six to nine months. When it came to selling these houses, the algorithm had them priced with growth based on certain assumptions that should have flipped them for profit. As these assumptions did not prove to be correct, it called for the execs to make tough decisions about proper pricing. Alarm bells did not ring out soon enough for Zillow as the sellers were taking the generous computer-generated offers at a much higher rate than expected. The conclusion was finally reached that Zillow had paid too much for their large inventory of houses.
Too much growth too soon: Zillow’s mantra was to gain on its competitor i-buying firms and take over as much market share as it could in as short a time as it could. But Opendoor and Offerpad, two other’s in the i-buying industry, took a more controlled approach to growth, and carefully refined their algorithms to reflect changes taking place in various markets. The real answer may never be known to the public, but somebody at Zillow may have been asleep at the switch when the computers were cranking out overly generous offers to homeowners for their property.
A parallel with the cruise business: A number of years ago I worked for a cruise line which had an automated booking system that tracked passengers from the time they selected the cruise until the cruise was finished. Typically, the pricing would be set in advance, and then discounted to fill empty berths in the ships. Somebody decided to discount several months of cruises all at once to bolster bookings and increase revenues. In a short time, these cruises were overbooked by up to 15%, and the passengers who were overbooked had to be offered cash incentives plus further discounts to move to a different cruise date.
Blame the computer: In the cruise line example, the execs tried to blame the computer system for allowing too many passengers to be booked. But computers are only as good as their programming, and their programming is based upon what the users want and/or need. The users of the data needed to comprehend what the information is telling them before they made decisions. As with Zillow’s example, somebody in an important position in the organization was not paying enough attention to the pricing process of the company’s product and wound up giving the store away.
Sources: Free exchange | Home-icide, The Economist, November 13th, 2021.
Thanks David. Even the pros make mistakes. Of course, a professional baseball player is considered a superstar even if he is only successful a third of the time at bat.