Skip to main content
Last Updated: 2025-09-19
Concept: Best Practices for Training Algorithms with Contract Clauses

Concept: Best Practices for Training Algorithms with Contract Clauses

Contract Management and Document Intelligence creates algorithms based on the contract clauses you create. We use the algorithms to better identify key phrases. Using the QuickAI feature, you can tag clauses and train models to recognize similar language across all your documents. You only need 1 example to get started, and adding more samples over time helps improve accuracy. The search feature lets you validate results and discover additional clauses to strengthen your model. These are a few best practices for training algorithms with contract clauses:
  • You can help Contract Management and Document Intelligence improve accuracy by submitting more examples, especially for clauses with varied wording in contracts. These examples show data privacy clauses can take different forms:
    • Example 1: Roles of the Parties and Compliance with Information Protection Laws
      GEM acts as the principal, and Supplier as its agent for collection, when using, processing, and disclosing confidential and personal information. Both parties must fulfill their respective legal obligations under applicable data privacy, information security, and breach notification laws (Information Protection Laws).
    • Example 2: Data Privacy
      Both parties must protect the personal data of each other's employees, customers, and suppliers. They must implement security measures, avoid third-party disclosure without written consent, and comply with applicable data privacy laws.
    When training a general algorithm to find all data privacy clauses, tag both examples and as many known variations as possible.
  • Tag clauses to train the algorithm: Tag all variations of a clause to ensure the algorithm recognizes different ways the clause can be written.
  • Use full sentences: Sample clauses must be complete sentences, without cut-off phrases, to ensure proper training.
  • Train 1 model per clause type: Training a single model for each clause type helps prevent confusion between similar clauses and improves tagging accuracy.
  • Leverage existing models when appropriate: Use existing Contract Management and Document Intelligence Search models to save time and take advantage of proven accuracy. Create a new model only if your clause contains company-specific language that isn’t covered by existing models.