Instant buying

Article summary

The Economist wrote about Zillow shutting down its instant buying business (i-buying), which used data-science to automate the process of buying and flipping houses. After much over-paying, Zillow expected to lose over $500 million and it's making 8,000 people redundant. Initially, this all may seem unexpected as house prices have risen 16-25% in the last year and a half. The article identifies three reasons why, for Zillow, i-buying was a stretch too far. The first reason was the anomalous state of the housing market during Covid. Volatile and unexpected prices do not play nicely with machine learning models trained on historical data. CEO Rich Barton said in some cases Zillow were listing houses for 6.2% less than they paid for them. However, companies like Opendoor were able to be pretty successful in this regard, on average offering home-owners an instant bid approximately 1.4% below market value. For Zillow, a second reason may be due to the fact that a small error in the model only has downsides. Home-owners are unlikely to sell for much less than what they think their property is worth - but will happily snap your hand off for a price that exceeds expectations. Indeed competitors like Opendoor and Offerpad seemed to be aware of this, making more conservative offers and not running into Zillows problems. Third, Zillow’s failure may have been strategic. Opendoor grew gradually and refined its algorithms, only entering six markets in three years. Zillow, desperate to play catchup, coughed up too much cash too many times.

What I thought

Programmatic buying is a fascinating industry, with the potential to disrupt in a similar vein to how programmatic ads revolutionised advertising and fuelled the rise of today’s biggest technology companies. Opendoor founder, Keith Rabois, clearly thinks so. Moving to a new arena in this space with his latest venture, Openstore - programmatically buying e-commerce stores. I think The Economist article touched on something that is true of many AI-fueled projects. When we bear in mind that 85% of AI projects fail to deliver ROI, it’s important to take heed of Tom Davenport’s recommendation - if you want to see the benefits of AI, forget about the moonshot. It seems like this is what happened with Zillow. They tried to do too much too quickly and their algorithms weren’t effective enough. For a business model that relies totally on accurate prediction-making, it pays to take time to ensure that the models are accurate before ramping it up.