Include routing when a meaningful subset can be handled more cheaply without violating acceptance rules. Measure the combined system, including router errors and fallback.
RouteLLM provides published evidence that routing can improve cost–quality trade-offs in its benchmark settings. It does not provide the buyer’s cost saving. A production test needs the expensive-call fraction, the number of double-processed requests, end-to-end latency and outcomes after escalation. Calibrate a routing threshold on development data and lock it before the final test.