The Cost of Accepted Work: Quality-Adjusted Economic Routing for Enterprise AI

Enterprise AI procurement and routing commonly observe the monetary cost of an inference call, whereas enterprises consume completed and accepted work. A low-price response may trigger verification, repair, escalation, retries, downstream correction, and governance effort that exceed the initial saving. Yet cost–quality routing, cascades, risk-aware selection, and abstention are already active research areas; consequently, the distinction between token price and economic value cannot itself establish novelty. This paper proposes Quality-Adjusted Economic Routing (QAER) as a falsifiable methodological specification for choosing among admissible model–verification–repair protocols according to the expected fully loaded cost of producing an independently accepted task outcome. It defines an acceptance contract, a task-episode cost ledger, hard privacy and risk constraints, terminal nonacceptance, estimation overhead, and a refusal action. The objective is a cost-per-accepted-outcome estimand evaluated with fixed coverage, safety, and workload controls, rather than an unconstrained ratio that can improve by dropping difficult tasks. We derive conditional economic comparisons, exhibit failure cases for token-cheapest selection, and propose a prospective, paired, leakage-resistant evaluation against fixed-model, classical cascade, benchmark-derived, and governance-aware routing baselines. No enterprise experiment, measured savings, causal effect, safety guarantee, or universal novelty claim is made. The scientific contribution sought is an inspectable experimental test of whether downstream acceptance economics changes routing decisions and improves total outcomes beyond strong existing methods after estimation and coordination expenses are charged.