Keep in mind that from an architecture perspective, the primary place to stop SQL injection attacks is by validating the input when it comes in. If the input doesn’t match appropriate patterns, especially in the case of a banking application where the likely patterns for each input should be easily defined, you reject it at that level. It then doesn’t get appended or inserted into a text string which becomes the SQL statement to be executed against a database server.
If you don’t get it at this level, the ability to prevent the SQL injection attack gets much harder. Perhaps IDS/IPS can detect based on some text matches. We might be able to do the same thing within the database, say by using DML triggers. However, if the appended text generates queries that are basically what normally gets sent back, none of the back-end solutions are going to be very effective.