so the modelling valuations reflect those kinds of vagaries, especially at the loan-level. zillow's model can predict a price within a given range with a pretty high degree of confidence. So assume your house has a Zestimate of $405,000; what that is telling you is that there is a 97% chance that the actual sales price of your home will be between $390,000 and $415,000. For modelling purposes, that actually makes a lot of sense, and for the large financial institutions who would use an automated valuation system, that's exactly what they're looking for.
But if you see that number as a consumer attached to your property, that means something entirely different. Should you price it at $385K? $425K? The difference of $25,000 is fairly significant. And that $25K range still puts you in the 97% confidence range. What if you are part of the 3% outside of that range? The value of a Zestimate in those cases is very different.
In short, it depends on who the end user of the AVM is going to be; if its a consumer, it's likely less valuable, because you want more of a bespoke, individual price assigned in order to maximize your value. But if you're a bank or investor looking to model out cash flow and predict default rates, and you need to know how secure the collateral is, then an AVM is probably much more useful and relatively inexpensive.