Use when the workflow is too slow, too expensive, or both and needs latency, cost, or token usage optimization.