Fraud detection is comprised of a lot of interesting data science and pattern recognition (example being utilization of graph databases like Neo4j and its advantages over traditional RDBMSs of index free adjacency) not to mention the leveraging of ML. If the business paying me to implement this sort of thing is 'stupid' and 'boring' (whatever your definition of that is in the context of a business), why would that make the act of working and implementing fraud prevention any less of an interesting endeavor?