Automating Invoice Classification with AI in Oracle

 

Classifying Invoices Using AI in Oracle Sustainability

Overview

Oracle Sustainability has introduced an advanced feature that leverages artificial intelligence (AI) to automatically classify invoice distributions not covered by predefined invoice classification rules. This enhancement reduces the need for manual review and simplifies the management of complex classification rules, improving efficiency and accuracy in sustainability ledger postings.

Previous Workflow

Before this update, invoice distributions that did not match any classification rules were processed as follows:

• Default Activity Creation: Oracle Sustainability generated activities using the activity type specified in the Default Activity Type for Invoice Classification profile option.
• Manual Review: These activities required manual review and classification before they could be posted to the sustainability ledger.
• Challenge: Achieving high classification coverage (i.e., a high percentage of correctly classified invoice distributions) often required creating increasingly complex rules, which were difficult to manage.

New AI-Powered Classification Capabilities

To address the challenges of manual classification and complex rule management, Oracle Sustainability now offers optional AI capabilities to automatically classify invoice distributions. These capabilities include:

Machine Learning

• Training Data: Utilizes a corpus of historical activities generated from Payables invoices already posted to the sustainability ledger.
• Model Attributes: The machine learning model analyzes a wide range of attributes from the invoice header, line, and distribution to identify patterns.
• Prediction Process:
  • The model predicts the most suitable activity classification category for an invoice distribution.
  • If the prediction is made with high confidence, an activity is automatically created using the predicted category.
• Benefit: Improves classification coverage without requiring additional complex rules.

Generative AI

• Use Case: Applied when there is insufficient historical data to train a machine learning model or when the model cannot predict with high confidence.
• Prediction Mechanism: Predicts the best matching activity type attribute combinations valid for the activity type specified in the Default Activity Type for Invoice Classification profile option.
• Example: If the default activity type is Purchased Goods and Services, generative AI predicts the most appropriate spend type.
• Input Fields:
  • Invoice line description
  • Invoice description
  • Item description
  • Purchasing Category and all levels of the Procurement Category Hierarchy (for purchase order-backed invoices)
  • Natural account description (for non-purchase order invoices)
• Benefit: Provides a fallback mechanism to ensure classification even in data-scarce scenarios.

Invoice Classification Process

The classification of each invoice distribution follows a structured sequence:

1. Exclusion Rules:
   • Evaluated first to determine if the invoice distribution should be excluded from further processing.
2. Classification Rules:
   • Evaluated in order of ranking.
   • If a rule matches, an activity is created with the specified activity type and activity type attributes.
3. AI Classification (Optional):
   • If opted into the AI feature, remaining unclassified invoice distributions are processed using AI.
   • High-Confidence Predictions:
     • Activities are created using the AI-predicted activity type and activity type attributes.
   • Low-Confidence Predictions or No AI Opt-In:
     • Activities are created using the Default Activity Type for Invoice Classification profile option.
     • These activities require manual review.
4. Outcome:
   • Classified activities (via rules or AI) are ready for posting, while unclassified or low-confidence activities need manual intervention.

Reviewing AI-Classified Activities

To ensure accuracy, Oracle recommends reviewing activities classified by AI before posting them to the sustainability ledger:

• Review Process:
  • Use a filter to view all activities classified by AI.
  • Verify the correctness of the activity type and attributes.
  • Make necessary corrections as needed.
• Continuous Improvement:
  • The machine learning model learns from corrected activities, improving prediction accuracy over time.
• Benefit: Combines automation with human oversight to maintain high-quality classifications.

Conclusion

The integration of AI into Oracle Sustainability’s invoice classification process significantly enhances efficiency by reducing manual reviews and simplifying rule management. By leveraging machine learning and generative AI, the system achieves higher classification coverage while maintaining flexibility for user oversight and continuous improvement.