AI Triage
How ProductSights automatically classifies, scores, and structures feedback.
Every piece of feedback that enters ProductSights is automatically processed by the AI triage pipeline. This happens within seconds of ingestion and requires no manual configuration.
What the pipeline does
For each incoming feedback item, the AI:
- Classifies it into one of six categories
- Analyzes sentiment (positive, negative, or neutral)
- Assigns a priority score from 0 to 100
- Extracts structured entities specific to the category
- Generates a summary of the original content
- Identifies product areas mentioned in the feedback
Categories
| Category | Description | Extracted Fields |
|---|---|---|
| Feature Request | A desired capability or enhancement | Desired outcome, use case, current workaround |
| Bug Report | Something broken or not working as expected | Steps to reproduce, expected vs. actual behavior |
| UX Issue | A usability pain point | Affected user flow, suggested improvement |
| Praise | Positive feedback about the product | What they like, perceived impact |
| Question | A question about functionality or usage | Topic area, question summary |
| Churn Signal | Indication of potential customer loss | Churn reason, competitor mentioned, risk level |
Priority scoring
The priority score (0–100) factors in:
- Urgency — how time-sensitive the feedback is
- Impact — how many users are likely affected
- Sentiment intensity — strength of positive or negative feeling
- Category weight — bug reports and churn signals typically score higher
Higher scores surface first in the insights list, helping your team focus on the most important items.
Triage statuses
Each insight moves through a status lifecycle:
- Pending — just ingested, awaiting AI processing
- Triaged — AI has processed and classified the item
- Approved — a team member has reviewed and approved it
- Rejected — a team member has decided it's not actionable
Viewing triage results
On any insight detail page, the AI Triage card shows:
- Category and subcategory
- Urgency level
- Product areas mentioned
- Source information
- Category-specific extracted entities (e.g., "steps to reproduce" for bug reports)
Smart deduplication
The triage pipeline also detects near-duplicate feedback using vector embeddings. When multiple customers report the same issue, ProductSights links them together and surfaces a Similar Insights section on each detail page, showing related feedback with similarity percentage scores.