Nebula AI Case Explorer™ | AI Relevancy and AI-Assisted Case Summary Revision

How it works

1. As you review documents in the Document Viewer, you will agree or disagree with the AI's relevance score, ideally with a reason (see Viewing Documents). Reasons are required for disagreements. 
2. The AI Relevancy tab shows a running picture of that feedback: how many documents have been reviewed, and your team’s agree/disagree split over time.
3. Once you've built up enough feedback, an admin can trigger a revision — the pipeline groups your disagreement records into patterns and proposes targeted additions or removals to your case summary. Select the ‘AI Analysis’ button to proceed.
4. You review the proposed changes and choose which to apply and which to skip. Each suggested change includes a confidence score and reason for the proposed change.
5. Once you are ready, you can move on to make manual edits to the case summary as you see fit.
6. You can use this revised case summary to generate an updated list of key themes. The key themes from your previous AI analysis version are available for you to compare to the updated key themes, and you can add previous key themes if you prefer.
7. As you proceed, we will provide a summary of the selections, then you can choose to generate an updated AI analysis based on the revised case summary and updated key themes.

When you're ready to trigger a revision

Three conditions make an AI-assisted case summary revision worthwhile:

• Clear pattern — your disagreements cluster around a recognizable topic, document type, or missing concept, rather than being scattered across unrelated documents.
• Substantive reasons — most of your disagreement records have a specific reason attached. A disagreement without a reason can't be used by the AI analysis.
• Broad enough sample — you've reviewed a reasonable spread of document types, not just one category, so the proposed changes aren't skewed.

Concretely: you need to disagree, with a reason, on at least 10 documents before the  Run AI Analysis action becomes available.

Writing good reasons

The quality of your reasons directly affects the quality of the proposed changes. Effective reasons:

• Name the specific topic or concept — e.g., "this document discusses SPE consolidation thresholds, central to the accounting fraud argument."
• Reference the document type where relevant — e.g., "board presentations about related-party transactions should be treated as always relevant."
• State the direction of the correction — e.g., "this is general HR correspondence with no financial content and should score lower."

Avoid vague assertions like "this is relevant" or "wrong score" — the pipeline needs specific, actionable reasons to produce useful proposals.

Reviewing proposed changes

When a revision completes, you'll see each proposed change — additions and removals to your relevance guidance — along with:

• Confidence scores for each proposed change with reason for proposal
• Any conflicting proposals, flagged for you to resolve manually

You can toggle individual changes off if you don't want them applied, then apply your selections and continue — you'll get another chance to edit before running the analysis. Nothing is forced through automatically.

After a revision

• Review proposed changes carefully before approving — a revised case summary can change scores on documents you've already reviewed.
• AI Case Explorer will run the AI analysis again using the revised case summary. Key documents, key people, key events, storylines, and document relevance scores may all change based on the revision.
• All versions of the case summary are retained, so you can roll back if needed — see Project List and Version History.
• If your team created notes, saved a document, or added or removed a document as Key, these changes will be preserved between versions.

Related articles

• Writing a Strong Case Summary
• Exploring Documents
• Project List and Version History