Skip to main content

How to extract data using AI Extractions

How to use AI Extractions in DataSnipper v26.1 to extract fields, tables, and footnotes from unstructured documents at scale using prompts and reusable templates.

AI Extractions automatically extracts fields, tables, and footnotes at scale from a wide range of document layouts. Every extracted value is traceable back to its source document, with reusable templates for recurring procedures.

Video tutorial

What the video covers

How to import source documents, set up AI Extractions, configure field types and keys, run the extraction, review results, and export to Excel.

Prerequisites

  • DataSnipper v26.1 or later

  • DataSnipper Accelerate or Elevate package

Steps

  1. Import your source documents (e.g. a set of payroll reports).

  2. Click AI Extractions in the DataSnipper ribbon and select the documents or folders to extract from.

  3. Set the extraction key (e.g. "Payroll Entries"). If the document contains list-based data, change the field type to List.

  4. Add each property to extract (e.g. Name, Hourly Rate, Gross Pay). Set the appropriate data type for each. Click Run.

  5. Review the extraction preview. If a field is not pulling from the correct location, add a short description to guide the AI, for example specifying which column a value should come from.

  6. Click Save as Template to reuse this configuration for future engagements.

  7. Choose your preferred export view and click Export to Excel. The extracted data and links to source documents are added to your workbook.

Working with templates

What the video covers

How to save an AI Extractions configuration as a template and apply a saved template to a new set of documents.

Field types

Text: use for names, references, descriptions, or inconsistently formatted dates.

Number (Decimal): use for monetary values, percentages, or rates where precision matters.

Integer: use for whole number counts or quantities where decimals are not relevant.

True/False: use for yes/no conditions such as checkboxes or validation confirmations.

Date: use for calendar dates. Where documents contain dates without a specified year and the format is ambiguous, use Text as a fallback.

Time: use for approval timestamps or other time-based values.

List: use for repeated data where structure may vary, such as registers or loosely structured tables.

Table: use for structured, repeatable tables with consistent rows and columns.

Did this answer your question?