3 ms·

As someone who doesn't work at Zillow but follows the space a little closer, don't buy into the superficial explanation/meme going around that they failed becau

As someone who doesn't work at Zillow but follows the space a little closer, don't buy into the superficial explanation/meme going around that they failed because of ML. My advice would be to read the earnings call transcripts over the last couple of years. What Zillow has communicated in the last earnings call is fairly contradictory to what they've been saying in previous calls. Some bullet points on my understanding of what went wrong: -They were paying way too much for too many homes. I believe this is ultimately due to internal incentives and ambitious goal setting. IIRC, they mentioned they had to alter their model output to be able to give themselves the permission to keep buying at high volumes. They lacked fiscal discipline here. -The holding times on these homes were too long, which increased their costs. My impression is that this was due to lack of operational experience with renovating and selling homes. Zillow was new to this, and they were trying to scale up very quickly, independent of whether they were efficient. -The core business model of Zillow historically has been as an advertising site/lead gen site. This business model is not in their DNA and their operational expertise. The negative margin pressure from this business in tougher times can really stress the overall business at times. Look at the other iBuyers in the space - they are still in the space. They just aren't as naive as Zillow was. Zillow's behavior here is really confusing and speaks to poor operations/strategy/discipline. Housing prices have been mostly been only going up in this market and Zillow was losing their shirt.

civilized 5y ago
Well, their strategy of blaming the nerds will at least ensure that they never have the opportunity to bet big on ML again. No one competent will want to work for them now.
- wussboy 5y ago
  I wish this was true but I suspect it is not. If Facebook can still have excellent engineers because they pay a ton of money, then so can Zillow.
  - ivanyahoocom 5y ago
    Zillow has always paid low compared to the market. They never got the “top nerds” and this was really a CEO problem. I would never bet on that company on any decently complex technology like predictive models. It took them years to make their model decent, something that a machine learning tutorial does better - just a linear regression on square feet and location. They were literally comparing my house to ones on the other side of the city - comparing a downtown property with outskirts.
    - selcuka 5y ago
      To be honest linear regression on square feet and location is what potential (human) buyers would mentally use, therefore it's a very, very accurate prediction of the price.
Spooky23 5y ago
In my area at least Zillow valuations really suck outside of subdivisions. They don’t do some fancy ML, just apply a factor with a cap to tax assessment based on comparable sales and listings. It mis-values houses near me by 20-40%. Other tools are much more accurate.

civilized 5y ago

Well, their strategy of blaming the nerds will at least ensure that they never have the opportunity to bet big on ML again. No one competent will want to work for them now.

wussboy 5y ago
I wish this was true but I suspect it is not. If Facebook can still have excellent engineers because they pay a ton of money, then so can Zillow.
- ivanyahoocom 5y ago
  Zillow has always paid low compared to the market. They never got the “top nerds” and this was really a CEO problem. I would never bet on that company on any decently complex technology like predictive models. It took them years to make their model decent, something that a machine learning tutorial does better - just a linear regression on square feet and location. They were literally comparing my house to ones on the other side of the city - comparing a downtown property with outskirts.
  - selcuka 5y ago
    To be honest linear regression on square feet and location is what potential (human) buyers would mentally use, therefore it's a very, very accurate prediction of the price.

Spooky23 5y ago

In my area at least Zillow valuations really suck outside of subdivisions. They don’t do some fancy ML, just apply a factor with a cap to tax assessment based on comparable sales and listings. It mis-values houses near me by 20-40%. Other tools are much more accurate.