Reading agent transcripts and escalations

Every conversation an agent handles is logged in full, whether it resolved on its own or got handed to a person — nothing about what happened to a customer's message is ever invisible to you.

Where to find them

AI assistant → [agent name] → Transcripts lists every conversation, filterable by whether it was Resolved by the agent alone, Escalated to a human, or Abandoned (the visitor left before an answer). Each entry opens the full back-and-forth, exactly as the customer saw it.

Why a conversation escalates

Every agent has an explicit confidence rule: if it is not confident enough in an answer, or the question touches money, cancellations, or clear frustration, it hands off rather than guessing. The transcript shows exactly where that handoff happened and why, so you can judge whether the rule is calibrated too cautiously or not cautiously enough for your business.

Using this to improve the agent

A pattern of escalations on the same type of question is the clearest signal that the agent needs more training on that specific topic — flag a few examples to us and we retrain against them directly, rather than guessing at what is missing.