Cost per customer
Last verified
Cost per customer is inference spend divided by tenant. In a product with a flat subscription and variable AI usage, it is the number that decides which customers are profitable.
Billing APIs cannot produce it — they do not know about your tenants. You need per-request tagging, which means a proxy or SDK instrumentation, or one API key per customer if the count is small.
Heavy users of an unlimited plan can be a real margin problem, and by the time it shows up in an aggregate bill it is a pricing problem rather than a monitoring one.