Hate it in concept (I can't speak to Pave specifically). Metrics are useful if the manager doesn't make hard targets, the data is good and the model is good. How do you model compensation packages? how does 20 days leave compare to unlimited leave? Are two companies with unlimited leave equivalent? When the managers have targets, they're incentivized to massage the data in their favor. So even if you're doing it right, that makes every other companies data suspect.

This also sounds like it will reinforce some negative behaviors. If the data shows that other companies are paying women 80% of their male peers, shouldn't this recommend I also pay women 80%? But I doubt the output will spell it out that way, will I even know it's influencing me this way?