Use customer signals to shape your AI agents

Tools for product managers to investigate customer feedback, build evals, and refine agent behavior.

Tuor trace review for the Returns assistant: a customer’s thumbs-down on an agent reply asking them to mail back a broken bowl, next to a thread where a policy expert says to refund orders under $50 with no return.

When customers hit problems with your agents, how quickly can you change their behavior?

Product managers should be able to see what went wrong and decide what should change, without waiting on engineering to dig through traces. Every delay means more customers hitting the same problem.

  1. Traces for the Returns assistant this week: thumbs-down, escalation and tool-failure counts with trends, above a list of recent customer conversations and their review status.
  2. Trace #8f3k2 being edited: the return-label tool call is swapped for a refund, and the agent’s reply is rewritten to say no return is needed, following the policy expert’s answer.
  3. The corrected trace being tagged eval-case from the tag picker, marking it as an expected example.
  4. Traces filtered by the eval-case tag with three selected, and the Share menu open on Create eval set.
Traces for the Returns assistant this week: thumbs-down, escalation and tool-failure counts with trends, above a list of recent customer conversations and their review status.

Take back ownership of the product experience