For an image to be given a label, the AI model must be more than 80% confident that the label represents what the image shows. The AI model is 91% accurate in labelling images of cars, 89% accuracy for vans and 82% accuracy for bikes which we aim to continually improve over time.