Managing Provision Models
Access the Provision Library by clicking the Provision Library tab in the left navigation bar.
The Provision Library lists all mapped and unmapped provision models. A provision model contains either a value (a date, a duration, a percentage, an amount of money, a pick list, or a short text) or language (text). Provision models are classified as built-in or custom. Mapped provision models can be applied when a document is imported for extraction. An unmapped provision model is one that has been detected but has not been selected for inclusion in extractions.
In addition to Discovery’s extensive library of built-in provision models, you can configure and train your own Custom Provision Models. You cannot edit or delete built-in provision models. You can, however, update them through syncing.
When the user reviews clauses from an extracted document, the on-screen highlight may not match information gathered correctly by the provision model. This issue affects only clauses, not fields. It is scheduled for repair.
Alert reviewers to this issue.
Viewing Provision Model Details
The Provision Library’s default layout provides information about stored provision models at a glance, including each model’s name, type, and last sync date.
Clicking a provision model’s name opens a detailed view of that model.
Reviewing Feedback
Review the quality of feedback from reviewers before the AI adopts it.
Use case: A Discovery administrator at an insurance company is training a custom Total Loss Determination provision and wants to make sure only high-quality reviewer input feeds the model. Checking the Feedback Approval tab from the Provision Library, the administrator surveys the default Document Types view and finds that a newly onboarded reviewer has been giving thumbs-up approvals without verifying them against the source document, a pattern visible when several of that reviewer’s approvals contradict the actual contract language. Rather than letting this feedback train the AI, the administrator opens the flagged document names to review the feedback in context, selects the reviewer’s incorrect entries, and rejects them. The correctly verified feedback from other reviewers is approved and used to train the model, keeping the provision’s accuracy from degrading because of one reviewer’s habits.
The same review-before-training discipline applies to any custom provision drawing on reviewer feedback, not only ones already showing an accuracy drop.
- Click the FEEDBACK APPROVAL tab. The Feedback Approval view opens, defaulting to the document type view.
- You can partially review a document’s extracted text by hovering your mouse over its Annotated Text field.
- Click the blue check mark to approve the feedback or the blue X to reject it. A confirmation pop-up message appears. Confirm your choice.
Syncing the Provision Library
Syncing the Provision Library brings in new provision models and updates existing ones from the AI service provider. It is a good idea to synchronize regularly to get new provision models from the publisher and benefit from ongoing improvements to the AI.
Use case: An administrator at a commercial insurance brokerage reads in Discovery’s latest release notes that Conga has added several new built-in provision models for cyber-liability coverage, which the brokerage has been tracking manually as a custom provision. Rather than waiting for the next scheduled maintenance window, the administrator opens the Provision Library from the left navigation bar and clicks SYNC LIBRARY, bringing the new cyber-liability provisions into the library alongside a quiet accuracy update to several existing built-in models. Because the sync can take a considerable amount of time, the administrator starts it at the end of the day so it completes overnight, well before reviewers need the library again the next morning. Regular syncing matters even without a specific new provision in mind, since it also picks up ongoing accuracy improvements to provisions already in use.
Analyzing and Reporting on Provision Models
You can review provision models in use and their effectiveness, and export detailed analytic reports.
- From Conga Start, click through .
- Click Provision Library on the left navigation bar.
- Click the Analytics tab.
- Filter provision analytics in the user interface by clicking the filter icon in the right navigation panel. You can filter by name, created date, modified date, source, type, language, accuracy, ID, or field ID.
- Sort filter results in ascending or descending order by clicking the Name, Source, Created Date, Type, Language, Source, or Accuracy column titles.