Training Custom Provisions
When custom provisions are newly introduced, the AI must be trained to recognize and extract them as described in their definition. Training is a user responsibility.
Training the AI consists of giving the AI samples, identifying the relevant contract terms, and directing the AI to the work. In practice, this involves finding the provision in the document, highlighting it, and selecting the appropriate provision-in-training. This occurs in the same context and workflow as the regular document review for built-in provisions.
Once an administrator has defined a custom provision and made it available, you can upload and import documents using the custom provision. On importation for extraction, CCI processes documents to extract the mapped provisions. For at least the first 30 contracts, and until the minimum accuracy threshold is met, users must review and highlight contract data that matches the custom provision.
To train a provision
A new provision remains in training for its first 30 reviews, and until the AI's results exceed the Minimum Accuracy Threshold set by the administrator.
Examples of Provisions under Training
The following examples show provisions highlighted for training.
Assigning a License Date Field
Assigning a Warranty Period
Assigning a Registration Number Field
Assigning a Fixed Fee Type Field
Assigning a Percent Price Increase Field
Assigning an Indexation Metric
Assigning an Indexation Review Point Field
Assigning an Arbitration Clause
Assigning a Picklist Value to a Provision under Training
Extracting from a Trained Custom Provision
A provision is trained when its F-score reaches the accuracy specified for the provision during its creation.
Once a custom provision is trained, it is available automatically for extraction on document import.
Retraining a Trained Custom Provision
Trained custom provisions are set to re-train continuously by default. You can modify or delete extractions from custom provisions and re-annotate the text if the text extraction is inaccurate. Your modifications to a trained custom provision's extractions retrain the custom provision.
If there are no extractions by a trained custom provision, you can add a new annotation to a custom provision and send it to the AI for training when your review is complete.