What is a confidence score?
A confidence score is a number an AI agent gives its own draft reply to show how sure it is that the reply is correct and safe to send.
A confidence score turns "is this draft good enough?" into something a rule can act on. A high score means the question was clear and the answer was well supported by the knowledge the agent had. A low score means something was uncertain: the question was ambiguous, the answer was not in the knowledge base, or the situation looked unusual.
What it is used for
In review, the score helps a person decide where to look closely. In automatic sending, it is the gate: drafts at or above a threshold are sent, and drafts below it are held for a person.
A confidence score is the model's own assessment, not a guarantee. It is one check among several, and it works best alongside a review period, send caps, and a threshold set cautiously at first.
Confidence scores in inboxr
inboxr scores drafts from 0 to 100. In autopilot, only drafts at or above the mailbox's confidence threshold are sent automatically, and the rest go to the review queue with the score shown. You can raise the threshold to be more cautious or lower it to send more. The analytics page reports average confidence over time.