When Valuation-Algorithms Go Wrong..? Lessons from ZILLOW OFFERS.

It will be fascinating to eventually dig into the underlying data and facts around the failures of Zillow's instant-buying (iBuying) division, known as - Zillow Offers. At a high-level, their valuation-algorithms and automated real-estate purchasing processing methodology LOST more than $420 MILLION (USD) in the just three months of July-September 2021.

That means that their Real-Estate Acquisition robots were LOSING an average of $6.7 MILLION (USD) on every single business-day in that quarter. How did the valuation-algorithms go wrong..? Why did no human catch the problem(s) earlier in the period..?

https://www.nytimes.com/2021/11/02/business/zillow-q3-earnings-home-flipping-ibuying.html

It is clear from some great reporting from BUSINESS INSIDER that "old-school" human real-estate agents across the country have reported cases of Zillow paying well above the expected market value of a home, which was combined with a?convenient chance for a homeowner to sell a home quickly directly to Zillow, as their CEO & Founder, Richard Barton, tried to quickly ramp up the iBuying home-flipping business to 5,000+ automated home-purchase transactions a month.

With 250+ high-dollar automated real estate purchase transactions every business day, it is very easy for a minor error in their machine-learning (ML) algorithms and robotic processes automation (RPA) methodology to reinforce their flawed home-price valuations - and instead of learning from their "mistakes" to infer that the incorrect-prices are both correct and the new standard-valuations to use for calculating future purchase offer-prices. [YIKES!]

It is important to note that Zillow was only offering this iBuying automated Zillow Offers program in approximately twenty-five (25) regional markets in the US, and just over one year ago - they were still working on an aggressive growth-strategy.

Now, only 13 months later, Zillow is sitting on thousands of houses worth less than what the company paid for them, and is looking to offload its remaining 7,000 houses to an unnamed buyer — or buyers — for $2.8 BILLION (USD)... which means the average property-price is $400,000 (USD).

It is one kind of corporate "mistake" when your machine-learning (ML) algorithms and artificial intelligence (AI) tools learn from the worst behaviours of Twitter users - like in the Microsoft example below. It is a much more serious corporate-oversight failure when your AI is conducting upwards of $100-MILLION (USD) per business day in residential real-estate purchases - and nobody at corporate HQ appears to have been "checking the math".

As the forensic-analysis of this Zillow Offers schmozzle becomes clearer, there will be some good "lessons learned" for other companies that are looking at machine-learning (ML) algorithms and robotic processes automation (RPA) methodology for their core-business models -

1. Did Zillow's property valuation algorithm fail everywhere a "little-bit" - or in just 1 or 2 regional-markets a LOT..?
2. What internal oversight was in place to look for "Rogue Robot" activity within Zillow Offers..?
3. Did Zillow's valuation algorithm & RPA "shift the market" in any regional-markets so much that it was creating an artificial feedback-loop that reinforced it's own incorrect high property valuations..?
4. Were Zillow Offers transactions being audited (and sanity-checked) against comparable non-RPA transactions in the same geographic markets..?
5. Was anybody HUMAN keeping an eye on the Robots..? Did they have any institutional power to raise their "Rogue Robot" concerns directly to the C-Suite executives at Zillow..?

I expect this Zillow Offers story might end-up as a future Michael Lewis book like "The Big Short " or "Flash Boys " - and when that happens it will be a book that I would probably read cover-to-cover in one sitting.

Artificial intelligence (AI) machine-learning (ML) algorithms and robotic processes automation (RPA) are all very valuable tools for corporate leaders to explore in their businesses - BUT just like any kind of corporate machinery you need to use them with "Due Care and Attention" and only after you understand the risks that occur from unsupervised use.