Self-learning AI in DII

The self-learning AI feature enhances the DII workflow by learning from user corrections made during invoice validation. When the RPA bot automatically loads invoice data and exceptions occur, for example, when Vendor, Company, or AOC values are missing or incorrect, the user manually updates the DII. These corrections are captured, stored in log tables, and used to improve future AI predictions.

Logging framework

Area Description
Log storage

The system maintains dedicated log tables to track extracted and edited values:

  • Document Interface Header Log
  • Document Interface Detail Log
  • Document Interface Distributions Log
  • Document Interface Add on Charges Log
Logging rules
  • OCR-extracted values are stored for reference.
  • Logs are created only when modifications occur.
  • When a new DII record is created, a Header Log record is automatically created.
  • When header data is updated, the same Header Log record is updated with the latest information.
  • When line level data is updated, the corresponding Detail, Distribution, and AOC Log record is updated.
  • If the record is interfaced without user edits, the initial Header Log record is deleted.
  • Once the AI learns from repeated user edits, future iterations automatically pick the correct value based on learned history.

Activation

Self-learning AI activates when user edits reach the hyper parameter threshold count defined during AI model deployment.

If no matching record is found using self-learned data, the system falls back to Normal AI logic.

See AI models.

This table shows the list of learning scenarios and inputs.

Learning type When learning is triggered? What AI captures? Inputs used
Vendor learning
  • Vendor missing from RPA extraction
  • Incorrect vendor selected by RPA bot
  • User manually corrects the vendor in DII
Final user-corrected vendor
  • Vendor Group
  • Extracted Vendor Remit To Address
  • Extracted Vendor Address
  • Extracted Vendor Name
Company learning
  • Company missing from RPA extraction
  • Incorrect company selected by RPA
  • User manually corrects the company
Corrected Company value
  • FEG
  • Extracted company name
  • Bill-to address
Add-on charge (AOC) code learning
  • AI does not identify the AOC code during processing
  • User enters the AOC Code field in DII
Final user-entered AOC code
  • Company
  • Vendor
  • Extracted description
  • Expense flag
    • If Expense is Yes, learning is stored in DIIDistributionsLog.
    • If Expense is No, learning is stored in DIIAddonChargeLog.