It's an incentives problem, not a technology problem. In the United States, interchange is one of the main sources of revenue both for card issuers (both debit and credit). Even though card network rails have their flaws (e.g. fraud, settlement speed, cost), it's unlikely that major banks will adopt another system unless it's as lucrative, or until there's regulatory action that forces them to do so.
Broadly speaking, what approach do you use to "build simpler 'explanation models'" from the more complicated "core fraud models"? Do you learn the models separately over the training data, or does the more complicated model somehow influence the training of the simpler model?