When you’re extracting metadata from documents with AI, a common question might be: How much should I trust the result?Confidence scores in Box AI Extract provide a probabilistic measure of extraction accuracy at the field level. Instead of treating every extracted value the same, confidence scores help you understand where the model is more confident—and where closer review may be needed.In this video, I show how confidence scores work with the /ai/extract_structured endpoint and what the scores in the response represent. Each confidence score is a value between 0 and 1 that estimates the likelihood that an extracted field value is correct.Confidence scores are designed to support human-in-the-loop workflows. They’re not guarantees, and even high-confidence extractions can be incorrect, but they provide useful visibility into model uncertainty so teams can prioritize review more effectively.If you’re building document extraction workflows with Box AI, this video walks through how confidence scores can help you decide when to trust automation—and when to take a closer look.

 137      3