Skip to main content
Version: 4

Best practices for Intelligent Document Processing accuracy

The accuracy of data extraction through Adeptia Connect's Intelligent Document Processing feature (Extract Data) is significantly influenced by the characteristics of the input file. Better accuracy is observed when the input file meets the following criteria:

  • Well-structured and labeled fields - PDFs or images with clearly defined and labeled fields enhance the extracted data accuracy.
  • Single or few records - Files containing fewer or single records simplify extraction and improve accuracy.
  • Fewer fields and data elements - Files with a limited number of fields and data elements reduce complexity and help you yield more accurate data.
  • Small file size - Smaller file sizes (in a few KBs) facilitate faster processing and improve data extraction accuracy.

Ways to further improve accuracy​

In addition to optimizing the input file for improved accuracy, you can do the following for even more accurate results:

  • Training the ML Model - Sometimes, the data extraction engine may struggle with non-text-based elements such as checkmarks or other types of selections. Adeptia provides a consulting service and can train the ML Model to work with such a file provided by the client.
  • Applying data mapping rules - In certain cases, applying custom data mapping rules as a post-processing step to the extracted output can help validate and refine the extracted data, ensuring accuracy from 96-98% to 100%.
  • Assigning clear names and descriptions to the output JSON fields - Assigning clear and descriptive names and accurate descriptions to the fields in the output JSON format enhances accuracy. This practice enables the AI to effectively map fields in the input file to the intended output fields.