I’ve been talking to teams running voice AI in production. And found a recurring pattern that the agent can be technically healthy, low latency, correct tool calls, good transcripts while bookings/conversions still drift down. The current workflow is manually reviewing calls, transcripts, objections, handoffs, lead quality, CTA timing, etc. For teams running thousands of calls, how are you actually diagnosing this today? Are you using structured event tagging, analytics, custom evaluation pipelines, human QA, or something else? Interested in people operating voice agents where conversion/booking rate is a real business KPI, rather than just demos.