That makes sense — tracking per-client usage becomes painful really fast once you have real traffic, especially with LLM + voice costs mixed together.
I actually ran into a similar thing from a different angle. What surprised me wasn’t only the accounting, but how operational behavior changes once multiple customers share the same underlying agent — the agent starts mixing assumptions unless the system keeps boundaries very explicit.
I’ll take a look at fluenty.io out of curiosity. I’m more interested in how you’re separating usage logically vs just billing — are you isolating sessions at the conversation level or at a longer customer lifecycle level?