Predictions in Information Extraction skill

  • Release version: Australia
  • Updated August 7, 2026
  • 1 minute to read

    • A prediction is the value that the Information Extraction skill infers for each field, table, or question in a use case. Predictions drive automation decisions and can be applied automatically or reviewed by a person, depending on the use case configuration.

    How predictions are generated

    A prediction is the value that the Information Extraction skill infers for a field, table, or question in a specific use case. When you submit a document for processing, the skill processes its content and generates a prediction for each field, table cell, and question set in that use case.

    A prediction shows what the skill identifies as the correct value based on the content in the document. Whether a prediction is applied without review or reviewed by a person depends on the automation setting for the use case. To understand how automation settings control whether predictions are applied automatically or sent for review, see Automation modes.

    Prediction sources

    Table 1. Prediction sources by use case element
    Use case element Prediction
    Field A single value recognized in the document, such as an invoice number or a date.
    Table Row-level values extracted from a tabular structure in the document, such as line items.
    Question An answer generated from the document content in response to a configured question.