With something as complex as employment decisions, there could be a lot of confounding factors. Example: What if A and B are located somewhere more desirable? There may naturally be a desire to stay where you are and less impetus to move. A and B might be in a class of company which confers more prestige, so there would be less impetus to move. A and B might entail more risk somehow. There are many of these.
Using bulk statistics to show things like bias and collusion is fraught with these issues.
If A and B are both located in desirable locations, and C is not, then the strangeness of there being more A->C->B jumps than A->B is only stronger.
Again, you're assuming that there only a few entirely orthogonal variables. If A and B are both located in a desirable location, the cost of moving to A or the cost of moving to B might be high enough, such that it's only justified if there is a large increase in salary. So this factor might well apply more to moving from C->A than it would A->B or B->A. The same housing expenses might also motivate other people to move from A->C. In this case, you might well see more A->C->B jumps than A->B.
However, none of these are cleanly applicable to real people in the real world. We're not dealing with particles or spherical cows here.
"It shows who to investigate," also implies a predetermined outcome and guilt. It's more that it shows there's something to investigate, period. In 2019, we should be cautious about applying bulk statistics instantly to blame or suspicion. It's that sort of attitude which propagates the 70% wage gap myth, when more detailed analyses narrow this to 94%.
Agreed. Such an approach would be a money wasting dragnet. Once upon a time the DEA tried a similar approach and began investigating people for buying bill counters (drug traffickers count a lot of bills). They eventually admitted it was a massive waste of time because without another reliable data source to correlate it with it just generated low quality leads that basically always led to nothing.
Using bulk statistics to show things like bias and collusion is fraught with these issues.