With a traditional bank you can walk into a branch and explain the situation to a human. I've had to do this several times with my large US bank, one time when traveling but thankfully to a place where they had a branch. In the end things got fixed up but not without much time and hassle expended.
I get the impression these institutions run ML software that was written by people who knew it would generate false positives but is now run by people who have been told that it is always correct. I say this having seen a bank manager call the group responsible on an internal phone line and them tell the manager that they suspect they (the branch manager) is a fraudster.
The big British banks are closing all their branches so that is becoming a thing of the past for us.
The legislation requires you to send any suspicious transactions to the regulator and requires you not to tell the customer.
So, from an ML perspective you want to maximize recall, precision be damned.