Every grounded turn has marginal COGS; at catalog scale naive RAG destroys margin. Leaders should tier retrieval, cache stable facts, and tie spend to revenue-risk classes—not model vanity.
Executive snapshot
The Token Tax: Why RAG for Retail is a Margin Game
Systems & logicToken economics
The move
Finance now asks for token and latency unit economics tied to categories and sessions.
The friction
More grounding often means more tokens and slower answers—checkout moments cannot absorb unbounded cost.
The product verdict
Implement budgets, routing, and fallbacks; measure $/successful intent, not tokens per curiosity.
Field note · Strategic dispatch · /dispatch/token-tax-rag-retail-margin