I agree that conversation analytics can be more actionable than dashboards alone. A resolution rate might tell you that something is working, but it doesn’t necessarily explain why customers are satisfied or frustrated. Topic trends and sentiment are especially useful because they can reveal gaps in documentation, recurring product issues, or prompts where the AI consistently struggles. I’d also pay close attention to thumbs down conversations and escalations they can provide some of the clearest signals for improving the agent. The key is turning those insights into a feedback loop. identify the recurring problem, improve the knowledge or workflow, then measure whether the conversations improve afterward. That seems much more valuable than simply watching support KPIs move up and down.