Our latest internal test: swapping a static FAQ block for an AI‑driven agent reduced first‑contact resolution time by 22% but spiked repeat tickets by 9%—users still ask the same question twice. Is the handoff timing or the agent’s context window the culprit?