How Zillow Lost $500 MILLION With Machine Learning

Zillow was leading the tech revolution in real estate. How could they make a $500,000,000 mistake in the hottest market in real estate history. Was this a problem with the Machine Learning Model they were using (the Zestimate), or something even more concerning? To be clear, the opinions expressed in this video are my own and based on what I think happened with Zillow. I only have access to public news articles and data, so this is all speculation. 0:00 Zillow Starting Strong 1:15 What is Zillow Offers? 2:37 $500,000,000 Loss 3:14 What is the Zestimate Really? 4:45 Is the Machine Learning Model at Fault? 7:15 How Did It Get Out of Hand? 9:55 Is Machine Learning Evil? In 2018 Zillow was riding high. They were dominating the real-estate marketplace with a digital tech offering. Their CEO, Rich Barton, had a wild idea. Could he use the estimates that they were generating to buy houses at under market value, fix them up, and then sell them for a profit. There were a lot of positives for this plan, first, home buying can be difficult. For the seller, buying directly through Zillow could be almost as simple as a click of a button, especially if they were getting a decent price for the house. You also could avoid many of the closing costs & additional fees with a third-party realtor. They already had a the zestimate system to evaluate housing prices. Could they just take that a step further and use that for real buying decisions? The company seemed to think so. That same year they started Zillow Offers, this platform would essentially make instantaneous offers on properties. This practice is known now as iBuying. By October of this year, Zillow disclosed that it wrote down over 500 million dollars of losses do to houses that were purchased at too high a price through the 3rd and 4th quarter of the year. Problems 1) The Zestimate wasn't used as it was initially intended. Did the company make the necessary changes to leverage it for buying houses? 2) House flippin