Build a PM workflow that auto-triages customer feedback

Description

You receive 500 pieces of customer feedback per week across Intercom, NPS, Slack, email — and your team reads maybe 30 of them. This builds an AI triage workflow that classifies, clusters, and surfaces signal so the PM only reviews what matters.

You are building an AI feedback triage workflow for {{team_name}}. Feedback sources: {{sources}}.

## Step 1 — Ingestion
Consolidate feedback into one system:
- Intercom conversations
- NPS/CSAT responses
- Support tickets (Zendesk, similar)
- Sales call transcripts
- Social (Twitter, LinkedIn) mentions
- Internal Slack #feedback-type channels

## Step 2 — Classification
Auto-tag each piece:
- Type: bug / feature request / praise / complaint / question
- Product area: feature or module
- Severity: low / med / high / critical
- Sentiment: negative / neutral / positive
- Customer tier: ICP / non-ICP / high-ARR
- Actionability: act now / watch / informational

## Step 3 — Clustering
- Weekly: cluster similar items into themes
- Rank themes by volume and severity
- Detect surfacing themes (new patterns vs. last month)

## Step 4 — Surface
| Signal | Channel | Cadence |
|--------|---------|---------|
| New theme emerging | PM Slack | Within 24h |
| Critical severity | PM pager | Immediate |
| Weekly roundup | PM email | Monday morning |
| Feature request themes | Roadmap board | Weekly |

## Step 5 — PM workflow
- 15 min/week on weekly roundup
- Drill down when a theme grows past threshold
- Ship decisions back into the triage system to close loop

## Output
1. Ingestion architecture
2. Classification taxonomy
3. Surface cadence
4. The one feedback type where AI classification is unreliable (needs human)
5. Expected hours saved per PM per week

Customize This Prompt