My general tip is to look for things that are almost but not quite automatable, where a human needs to do a quick look-over. One big example is fraud: it's nex

My general tip is to look for things that are almost but not quite automatable, where a human needs to do a quick look-over. One big example is fraud: it's next-to-impossible to define a 100% accurate set of rules to filter fraud, but it's often easy to train an algorithm to catch the worst offenders, or flag suspicious cases to significantly narrow the amount a human needs to review.