Are there any widely available, published threat models for secret management of this sort? It seems to me that if you want to frame the discussion in a useful way, you'd start there. Then you can point out where current practice works, where it doesn't, and how your solution works better.