Don't head straight for a machine learning solution, first deploy something simple.
Eugene Yan writes some of the best articles about ML. His latest advises that the best place to start a machine learning type project is to not use ML.
Machine learning evolved as an alternative way to solve the sorts of problems computers traditionally found hard - finding and coding complex relationships in data that were hard to write down and maintain by hand.
Because of successes in some fields, notably vision, people have begun to use ML for all sorts of other tasks. This can also be successful, but it does come with many overheads, particularly in building, maintaining and scaling complex model pipelines.
Yan's suggestion is that you should always start with something simple. I agree with this. If the problem can partially be solved either by manual labelling or by some simple rules, then do that. Even if you progress to an ML solution later, putting in place a simple solution helps you to get to know the problem from end-to-end.