I've found the MLS algorithms to be interesting too.

The zestimate is basically a computer algorithm, it doesn't know the condition, features, etc of the house and can only guess the value based on the information the computer gives it. 

They have the algorithms to assess thousands and thousands of properties a day.

Optimization / Targeting: In essence, this is you telling Facebook's algorithm what your goal is, for example you can tell it to get you clicks to your website, or even get people to click into a lead gen form where you capture their information to reach out personally or via email. 

They say one team was able to put together an algorithm which made the Zestimate very accurate with only a 2.5% margin of error.

If you try to avoid this inquiry by depositing the money systematically in smaller amounts, that is called "structuring" and they have algorithms to detect this behavior and then they will really want to know the deal. again, if you're doing everything legally, then go to a bank and deposit your money and answer their questions honestly.

However, the one aspect that did strike me as unique with them is that they approach data collection from an algorithmic direction versus a call and survey approach like Costar.

Airbnb has continued to tweak their search algorithm, and new listings now get lots of juice in the first 3 months.

Those values are based off an algorithm which no one really knows. 

FICO's exact algorithms are a trade secret, things that "should" bump someone's FICO often do not pan out, so you try these things BEFORE they even start looking at houses.