When a Kuwait address read should page a person

The job is a photo the reader already tried. The votes do not agree, or the string will not match the ticket. Someone still lets the driver guess, or a reviewer finds the pile at the end of the shift.

Teams ask for “better OCR.” That is a model. This loop pages a person when confidence is below the line you already wrote, holds the stop, and stops inventing an address.

What the automation actually does

It reads the vote and the match result the ensemble already produced. It holds the delivery when the score is below your threshold or the string misses the stored address. It sends one note with the photo, the ticket, and the conflicting reads. It does not rewrite the plate.

FleetOptics address detection is that ensemble: more than one reader, a vote, then a check against company records. The leftover is this queue — the shots a single pass should not close. AI automation is the hold-and-page. We will not treat a thumbs-up in the driver chat as a matched address.

This is not deciding whether field photos need a reader. That call scores your captures and asks if a VA queue is still cheaper. This loop assumes the reader already runs. It is also not a proof photo to close or a pickup gate. Close argues the door. Pickup argues the bag. This job argues a string the cameras could not settle.

When we leave the photo with a person

If ops still marks delivered when the plate is unreadable, software should not pretend a threshold exists. If the useful match is a WhatsApp voice note from the rider, write the expected address into the ticket first. If a manager can override every miss, log the override — do not hide it.

Kuwait tickets often use Area/Block the map does not know. That is a match rule, not a reason to skip the page when the readers disagree.

Bring twenty misses from last week and the address fields you already store. Book thirty minutes or write hello@jamilglobal.com.