A hotel booking agent shows a high action count. Traces show the booking tool failing and the agent retrying the same call. Which metric should operations use to judge this agent?
Select an answer to reveal the explanation.
Short Explanation
Redialing a busy number all afternoon is not the same as holding a room. You measure bookings the agent actually completes, not the tool attempts inflated by retries. A rising attempt count can mean the booking tool is failing.
Full Explanation
Action accuracy for this agent is a booking that was created, not a call that was tried. The traces show the tool failing and the agent retrying, so an attempt count climbs while guests still have no reservation. Counting each retry as another success turns the outage into a good day on the report. Average reply length can be long on a failure message and short on a completed booking, so it does not tell operations whether a room was held. Reporting the retry count as proof of a healthy path confuses persistence with completion, and a rising retry line is the failure signal, not the success signal. Successful bookings stay comparable across days even when the tool is noisy, and the attempt count can sit beside them as a diagnostic rather than as the score. Exam caveat: retries can be a legitimate reliability tactic, and they still must not be added into the success metric. Operational check: fail the booking tool in a test, confirm the attempt count rises, confirm the completed-booking count does not rise with it, and show both numbers together so retries cannot be read as bookings.