Reducing manual inspection costs through object detection in delivery completion photos - Kurly Tech Blog

Kurly automated manual delivery-photo inspection by training an object detection model to spot purple boxes and paper bags in completed delivery images.

Using a data-centric approach with YOLOv11 and Ultralytics, the company improved detection quality and reduced review workload.

Kurly then expanded the model to detect vinyl-state deliveries as well, cutting review targets and errors even further.

The effort reduced manual inspection costs by 93%, and later achieved a 99.8% reduction while improving service quality.