Confidence & escalations
The confidence threshold
One number decides what sends alone and what waits for you. The line is always yours.
How the score works
Every draft gets a confidence score from 0 to 1, reflecting how well the answer is grounded: did the AI find the order, does a knowledge document back the claim, has this question been resolved the same way before.
Scores land in three bands everywhere in the app: high is 0.80 and up, medium from 0.60, low below that.
Three outcomes
Above your threshold: send_now, the reply goes out on its own. Below it: queue_and_wait, the draft parks in your approval queue. Certain signals skip the math entirely and escalate straight to a human.
Choosing your number
Most merchants start at 0.90 for two weeks, review what auto-sent on the dashboard, then lower it as trust builds. The threshold is per-workspace and changes apply to new drafts only.