AI-Powered Similar Cases Adds Vector-Similarity Search to CRM Case Records

What changed

A new Similar Cases tab has been added to the case-summary popup (the one triggered by Generate Insight on a case record). Under the hood, NetSuite feeds five case fields — subject, first message, item, type, and issue — into an AI model that performs vector-similarity comparison against an index of closed cases. The feature returns up to five results ranked by a percentage-based similarity score.

Indexing details

• Eligibility cutoff: Only cases whose status changed to closed after 7 October 2026 are added to the index. Pre-existing closed cases are not retroactively indexed.
• Index cap: The system considers up to 10,000 of the most recent closed cases. The release notes do not mention a way to raise this limit or whether it is per-subsidiary.
• Latency: Bulk imports of closed cases (e.g., CSV import) may take several hours to finish indexing before those cases appear in similarity results.
• Open cases are excluded from the similarity pool entirely.

Data handling

Oracle states the five fields sent to the AI model never leave Oracle infrastructure and are not used for model training. There is no opt-out mechanism described beyond disabling the Narrative Insights preference.

Prerequisites

1. CRM feature flag: Setup > Company > Enable Features > CRM subtab > Customer Relationship Management must be checked.
2. Narrative Insights preference: The Enable Narrative Insights toggle on the AI Preferences page (Narrative Insights tab) must be on. It is enabled by default in new accounts.
3. Permission: Users need at least View level on the Support Cases permission to access the Similar Cases tab.

What to do

1. Review your AI governance posture. If your organization restricts sending case data to AI models — even within Oracle infrastructure — consider disabling the Narrative Insights preference before the feature activates. The preference is on by default.
2. Understand the indexing gap. Because only cases closed after 7 October 2026 are indexed, the similarity pool will be sparse initially. If you need historical coverage, the release notes offer no backfill mechanism — confirm with NetSuite Support whether one exists.
3. Account for the 10,000-case cap. High-volume support teams may exceed this quickly. The release notes do not clarify whether the cap is configurable, how oldest cases are evicted, or whether the cap is account-wide or per subsidiary. Verify with Oracle if this limit affects your use case.
4. No SuiteScript or REST API surface exposed (yet). The feature is UI-only via Generate Insight. There is no mention of new records, saved-search columns, SuiteScript modules, or REST endpoints for programmatic access to similarity scores. If you need similarity data in workflows or integrations, you cannot automate this today.
5. Thumbs up/down feedback loop. The UI includes a feedback widget at the bottom of the Similar Cases tab. Let your support team know this exists — Oracle likely uses it to tune relevance ranking, so early feedback may improve results faster.
6. Bulk import timing. If you routinely import resolved cases (e.g., from a migration or external helpdesk sync), plan for a multi-hour indexing delay before those cases surface in similarity results.

Open questions

• Is the 10,000-case index limit configurable, and how does eviction work when the limit is reached?
• Can the similarity data (scores, matched case IDs) be accessed via SuiteScript, SuiteAnalytics, or SuiteTalk in a future release?
• Is there a way to backfill cases closed before the 7 October 2026 cutoff?
• Does the index respect role-based case access restrictions, or can a user see similarity matches for cases they would not otherwise have permission to view?