As a REALTOR and an economist I don't think it is that accurate Steve.
There are 4 methods to value a property in Real Estate. The 4th or Multiple Regression Analysis is probably most similar to what zestimate does. It essentially is a statistical method for spitting out a price based on several factors, once enough data points are obtained. But frankly what zestimate does is oversimplify this process and the more they do that, the less accurate it is.
When an appraiser uses this method (or an economist) there are literally over 100 inputs. Zestimate doesn't have anything close to this, they might have 2-3 dozen at best (based on a very thorough assessor's data) and this too is likely to be quite outdated. Think about how long it takes the assessor to figure out that you put new vinyl siding on your house, or added some skylights, etc.
The point is, that to do something like this is very complex and sure a computer program can (and is required) to do it. But this only really works if they have those 100+ data points. Everything from the type of moldings and floor coverings to a time dating variable (to account for inflation, appreciation, depreciation, etc. in a given market) are considered. How could zestimate possibly do this?? They cannot.
So ultimately what you have is a "zestimate", just that something close, maybe within 10% of the price. And frankly that's about as accurate as what your neighbor Bob, or your cousin Sherrie could tell you.
If you want a good value, call a couple REALTORS, they'll typically do a comparative market analysis for free and I think this is generally a better method, at least on residential real estate.
Broker, CRS, GRI, ePro
Raving Real Estate