voice-of-customer-synthesizer（客户声音合成器）

Voice of Customer Synthesizer

Turn scattered customer feedback into a single source of truth. Aggregates signals from every source you have, clusters them into themes, and produces a report that product, marketing, and CS teams can actually act on.

Built for: Startups where customer feedback lives in 6 different places and nobody has time to synthesize it. The founder says "what are customers saying?" and nobody has a clear answer. This skill produces that answer.

When to Use

• "What are our customers saying?"
• "Synthesize customer feedback from last quarter"
• "Build a VoC report for the product team"
• "What themes are coming up in customer feedback?"
• "Aggregate feedback from all our channels"

Phase 0: Intake

Feedback Sources (provide all you have)

1. Support tickets — Export from support tool (CSV: customer, date, subject, description, tags, resolution)
2. NPS/CSAT survey responses — Scores + verbatim comments
3. Slack messages — Customer channel messages, feedback channels
4. G2/Capterra reviews — Will scrape if product is listed (provide product name or URL)
5. Call/meeting transcripts — Customer call recordings or notes
6. Churn exit survey responses — Why did customers leave?
7. Feature request log — Internal tracker of what customers have asked for
8. Social mentions — Twitter/LinkedIn/Reddit threads mentioning your product
9. Email threads — Notable customer emails (praise or complaints)
10. In-app feedback — Any in-product feedback submissions

Configuration

1. Time period — What window to analyze? (Last 30 days, quarter, 6 months)
2. Product name — For review scraping and context
3. Report audience — Who's reading this? (Product team, exec team, CS team, all)
4. Focus areas — Any specific themes to pay attention to? (e.g., "onboarding experience", "pricing feedback", "mobile app")

Phase 1: Data Collection

1A: Internal Data Processing

From the provided inputs, normalize all feedback into a standard format:

SOURCE | DATE | CUSTOMER | SEGMENT | FEEDBACK_TEXT | SENTIMENT | CATEGORY

Sentiment classification per item:

• Positive — Praise, satisfaction, delight
• Neutral — Feature request, question, observation
• Negative — Complaint, frustration, disappointment
• Critical — Churn threat, escalation, anger

1B: External Review Scraping (if applicable)

If product is on review platforms:

Chain: review-scraper for G2, Capterra, Trustpilot
Filter: reviews from the target time period

Extract: rating, review text, reviewer role/company size, date, pros, cons.

1C: Social Listening (if applicable)

Search: "[product name]" feedback OR review OR "switched to" OR "stopped using"
Search: "[product name]" site:reddit.com OR site:twitter.com

Phase 2: Theme Clustering

Group all feedback items into themes using a bottom-up approach:

Clustering Method

1. Read all feedback items
2. Identify recurring topics (mentioned by 3+ customers or in 3+ sources)
3. Group into theme clusters
4. Rank by frequency AND severity

Theme Template

THEME: [Name — e.g., "Onboarding Complexity"]
FREQUENCY: [N mentions across M sources]
SENTIMENT: [Predominantly positive/neutral/negative]
TREND: [↑ Growing / → Stable / ↓ Declining vs prior period]

REPRESENTATIVE QUOTES:
- "[Exact quote]" — [Source, Customer segment, Date]
- "[Exact quote]" — [Source, Customer segment, Date]
- "[Exact quote]" — [Source, Customer segment, Date]

CUSTOMER SEGMENTS AFFECTED:
- [Segment 1: e.g., "New customers in first 30 days"]
- [Segment 2: e.g., "Enterprise accounts"]

ROOT CAUSE HYPOTHESIS:
[1-2 sentences: Why is this coming up? What's the underlying issue?]

IMPACT:
- On retention: [High/Medium/Low]
- On expansion: [High/Medium/Low]
- On acquisition: [High/Medium/Low]

Phase 3: Analysis

3A: Sentiment Overview

Overall Sentiment Distribution:
  Positive:  [N] items ([X%])  ████████░░
  Neutral:   [N] items ([X%])  ████░░░░░░
  Negative:  [N] items ([X%])  ██░░░░░░░░
  Critical:  [N] items ([X%])  █░░░░░░░░░

3B: Source Comparison

Insight: Different sources often reveal different stories. Support tickets skew negative (problems). Reviews skew bipolar (love/hate). Calls reveal nuance. Note where themes appear across sources for highest confidence.

3C: Segment Analysis

3D: Trend Detection

Compare against prior period (if available):

New themes this period: [Themes that weren't present before] Resolved themes: [Themes that decreased significantly — things you fixed]

Phase 4: Recommendations

For Product Team

For CS/Support Team

For Marketing Team

Phase 5: Output Format

# Voice of Customer Report — [Period]
Sources analyzed: [list]
Total feedback items: [N]
Date range: [start] — [end]

---

## Executive Summary

[3-5 sentences: What are customers saying? What's the overall sentiment?
What's the single most important thing to act on?]

---

## Sentiment Overview

Positive: [X%] | Neutral: [X%] | Negative: [X%] | Critical: [X%]

Net Sentiment Score: [calculated — % positive minus % negative]
vs Prior Period: [+/- X points]

---

## Top Themes (Ranked by Impact)

### 1. [Theme Name] — [Sentiment] — [N mentions]
**Summary:** [2-3 sentences]
**Key quotes:**
> "[Quote]" — [Source]
> "[Quote]" — [Source]
**Recommended action:** [What to do]
**Owner:** [Product / CS / Marketing]

### 2. [Theme Name] — ...

### 3. [Theme Name] — ...

[Continue for top 5-8 themes]

---

## What Customers Love (Preserve These)

| Strength | Evidence | Marketing Opportunity |
|----------|---------|----------------------|
| [Feature/experience] | "[Quote]" — [N mentions] | [How to use in messaging] |

---

## What Customers Want (Feature Requests)

| Request | Frequency | Segments | Product Priority |
|---------|-----------|----------|-----------------|
| [Feature] | [N mentions] | [Who wants it] | [P0/P1/P2] |

---

## What Causes Pain (Fix These)

| Pain Point | Severity | Churn Risk | Recommended Fix |
|-----------|----------|------------|----------------|
| [Issue] | [High/Med/Low] | [Yes/No] | [Action] |

---

## Trends vs Prior Period

[What's getting better, what's getting worse, what's new]

---

## Team-Specific Action Items

### Product Team
1. [Action] — [Evidence]

### CS Team
1. [Action] — [Evidence]

### Marketing Team
1. [Action] — [Evidence]

---

## Appendix: All Themes Detail

[Full theme cards with all quotes and analysis]

Save to clients/<client-name>/customer-success/voc/voc-report-[YYYY-MM-DD].md.

Scheduling

Run monthly or quarterly:

0 8 1 */3 * python3 run_skill.py voice-of-customer-synthesizer --client <client-name>

Cost

Tools Required

• Optional: review-scraper for G2/Capterra/Trustpilot reviews
• Optional: twitter-scraper for social mentions
• Optional: reddit-scraper for community feedback
• All analysis is pure LLM reasoning on provided data

Trigger Phrases

• "What are customers saying?"
• "Build a VoC report"
• "Synthesize our customer feedback"
• "Run voice of customer analysis"
• "Customer feedback summary for [period]"