For large N, that could make some sense. For a lot of (company, job) pairs, though, that'll be an N of 1. That seems unrealistic. Of course we can determine

For large N, that could make some sense. For a lot of (company, job) pairs, though, that'll be an N of 1. That seems unrealistic. Of course we can determine some threshold below which it doesn't apply, but then there are games to be played around how jobs are grouped, etc. I'm not saying there's no solution down that path, but I don't think it's "easy".