As someone who works in the (Australian) industry and is familiar with the various sorts of rules engines and automated valuation models (AVMs) in use, it's really not that hard to predict the selling price of a *standard, boring suburban house* so long as you have a rough idea of the land area, living area, age, bedrooms, bathrooms, etc.
You then index transactions/settlements on other properties with similar attributes and get a range of likely values. When the figure comes back "spot on", it's more like that behind the scenes. Feeds of these transactions are published fairly widely, and it's easy to just index everything "just in case" something in the area transacts.
Anywho:
1) Your loan to value ratio was beneath a certain level of risk for the bank
2) The range of values out of the automated model were considered accurate enough
3) Its often easier for a set of rules around lending decisions to mark you as 'acceptable risk' and display a matching figure to an end user/decision maker than to display a range or something a few hundred dollars off *if* you pose no obvious risk. Either that, or you had a contract of sale, so the valuation is only really agreeing your offer is reasonable.
Where it gets interesting is when the housing is non standard - proximity to waterfront views, industry, rail, high value, weird zoning, etc - if there was anything like that flagged, people would be involved to get the 'ground truth'.
