Ironclad Feature Review: Best Practices for AI Clauses
We’ve been delving into Ironclad’s AI Playbooks since its launch, exploring its potential to streamline contract reviews by flagging agreements that deviate from your standard provisions. We’ve identified ways to make the most of recent enhancements to the feature:
#1: Augment Ironclad’s Standard AI Clauses
Ironclad currently offers 175 out-of-the-box AI-detected clauses and now allows you to enhance the AI detection capabilities by training the model with additional real-world examples. We believe this enhancement is crucial. Ironclad suggests training with one to three examples per clause. We found training with 10 real-world examples from various use cases, contract types, and industries resulted in the most helpful AI detection. We suggest examples taken from your own repository and GPT models.
#2: For Ironclad Customers in Unique Verticals: Use Custom AI Clauses
Ironclad’s 175 out-of-the-box AI-detected clauses cover common commercial terms. But Customers with specific review needs can use the same training capabilities discussed above to ensure their clauses can be included in an AI Playbook. Because Custom AI Clauses are developed on a customer-by-customer basis, we recommend using the 10 examples to train a Custom AI Clause model.
#3: Use New Training Impact Feature for Better AI Training
Ironclad now assesses the effectiveness of the training examples you upload when creating a custom AI-detected clause. Based on this indicator, you can decide whether or not to keep the example. Keep in mind, the impact rating relies on the initial example you provide, so start with the clauses that are the best representations of their type.
#4: Adjust Similarity Thresholds to Align Clause Detection Settings with Your Goals
Control the sensitivity of AI detection for clauses processed through AI Playbooks and Smart Import. You can opt for an exact match, highly similar, or somewhat similar. We agree that Ironclad’s recommendation of “highly similar” is the best setting to apply to Standard AI Clauses. For common and generic clauses, “exact match” could result in missed detections and “somewhat similar” could cast too wide a net. For Custom AI Clauses, we recommend “somewhat similar.” This setting acts as an additional validation layer as you refine the training of a custom clause.
Ironclad’s AI capabilities are rapidly improving. Our goal is to partner with our clients to bring them the benefits of Ironclad’s latest product advancements. We’d be happy to discuss how we can assist you and your team.