* She also said, "I think banks are smart enough not to use actual Zip Code information to decide who to give a loan to, because that's just so obvious, but most data-science algorithms do use geolocation information to decide, for example, who is a 'high-value' customer." I'm curious now what information that might be.
Well, they might just convert you address into actual geocoordinates, which they then feed into a fancy algorithm nobody really understands. One problem with some machine learning algorithms is low explainability: They can have high accuracy in predicting what they are supposed to predict, but are hard to understands in terms of which parameters are decisive. But if you want to give advantages to the rich people (because that is better for business) without explicitly saying so, an opaque algorithm might be exactly what you want.
And ZIP codes have the disadvantage that the boundaries are somewhat arbitrary, so if you switch to something else, you at least are not screwed over by living on the wrong side of an imaginary line.
But even if explicit geoinformation is included, it might be inferred from other data. For example, the school you went to probably tells a lot about what kind of neighborhood you grew up in. Or your neighbors could be inferred from your contact list on social networks and so on.