We will soon release confidence levels and human-in-the-loop review in Box Preview, a powerful new capability that brings transparency, control, and trust to AI-powered data extraction in Box Extract. With this capability, users will be able to view how confident the Box Extract Agent is with every extracted metadata value, enabling them to validate whether the extracted value is correct or whether it should be modified all without leaving Box Preview.
This feature will apply specifically to metadata generated through the autofill capability within in Box Preview and Box Apps, as well as files processed using Custom Extract Agents in Box Extract. Every metadata value extracted using autofill or within Box Extract will include a low or high confidence level, or a model-generated probability indicating how likely the extracted value is correct. Confidence level will be displayed directly in the metadata panel next to the corresponding metadata value.
Fields that contain low confidence level will be visually flagged in the metadata panel, making it easy to identify and review incorrect metadata without manually scanning every value.
For fields flagged for review, reviewers will have two simple inline actions directly in Box Preview:
- Accept: Confirms the extracted value is correct. The field is marked "Accepted", the confidence flag is cleared, and the value is immediately saved.
- Edit or Delete: Requires the user to enter the correct value. The field is marked "Edited," and both the confidence level and bounding box are removed.
With this release, Box Extract will provide customers with the confidence they need for data they can trust.
Stay tuned to learn more about this release.