I used to work at a trading systems vendor and once in a while, someone would ask "couldn't we use ML to tell our clients what to buy and how much to pay?" My r

I used to work at a trading systems vendor and once in a while, someone would ask "couldn't we use ML to tell our clients what to buy and how much to pay?" My response was that "if we could do that, we'd never give it to our customers and just trade ourselves." So I always take things like the Zillow estimate with a grain of salt because "if Zillow really believed this number, they'd use it to buy houses rather than put it on their website." So it was interesting when Zillow actually started to essentially "trade on their numbers" while also sharing them with the public. That was a very bullish sign that the company deeply believed the quality of its signals. That can be totally normal. Every investment (and in fact, every project - which is a form of investment) happens because people believe that it's a good bet which is different than guaranteed. Some of the most money-making hedge funds are right in only like 51% of their bets, but that's enough! The thing that somehow went wrong is simply risk management. It seems like nobody with power at Zillow asked the question "how do we make sure this doesn't ruin the company if it's wrong?" UNLESS the scenario we're seeing now is their predicted acceptable worst case scenario. It's possible that they simply said "this is a juicy bet we want to make. If it goes wrong, we'll loose X millions of dollars and have to fire 25% of our team. Are we willing to take this bet?"