synthesizing-voc

name	synthesizing-voc
description	Use when you have customer feedback files (transcripts, tickets, surveys) and need to identify themes and opportunities without hallucinating.

Synthesizing VOC

Overview

Converts raw Voice of Customer data into decision-grade themes and opportunities. Strict protocol prevents hallucination by requiring verbatim quotes, source attribution, and minimum source thresholds.

When to Use

• Have 3+ interview transcripts, support tickets, or survey responses
• Need to understand "what are customers actually saying?"
• Preparing for roadmap planning or prioritization
• Want to validate assumptions with real customer evidence

Core Pattern

Step 1: Validate Sources

1. List all files in inputs/voc/
2. Count total sources
3. If fewer than 3 sources:
   • State: "Insufficient sources (found N, need 3+)"
   • List what you found
   • STOP and ask user for more data
4. Read each file completely

Step 2: Tag Verbatims (Internal)

Create a mental index:

• Verbatim Quote → Source File → User/Segment (if known)

DO NOT summarize or paraphrase at this stage.

Step 3: Identify Themes

Find patterns appearing in 3+ separate sources. For each theme:

Required	Rule
Core pain point	One sentence, factual
Frequency	Exact count of sources
Quotes	Up to 3 verbatim (if fewer, say so)
Segment	Only if explicitly stated
Impact	Only if user explicitly stated

Step 4: Generate Output

Write to outputs/insights/voc-synthesis-YYYY-MM-DD.md:

---
generated: YYYY-MM-DD HH:MM
skill: synthesizing-voc
sources:
  - inputs/voc/interview-1.md (modified: YYYY-MM-DD)
  - inputs/voc/interview-2.md (modified: YYYY-MM-DD)
downstream:
  - outputs/roadmap/Qx-YYYY-charters.md
---

# VOC Synthesis: [Date]

## Summary
[2-3 sentences: what did we learn?]

## Themes

| Theme | Frequency | Segment | Impact | Confidence |
|-------|-----------|---------|--------|------------|
| [Theme 1] | N sources | Explicit/Unknown | Explicit/Unknown | High/Med/Low |
| ... | ... | ... | ... | ... |

## Voice of the Customer

### Theme 1: [Name]
**Pain point:** [One sentence]

> "[Verbatim quote 1]"
> — Source: [filename], [user/segment if known]

> "[Verbatim quote 2]"
> — Source: [filename]

### Theme 2: [Name]
...

## Top 5 Opportunities
1. **[Opportunity]** — Evidence: [N sources mention this]
2. ...

## Top 5 Terminology Gaps / Misconceptions
1. **[Term/Concept]** — What users say vs. what we mean
2. ...

## What to Validate Next Week
- [ ] [Specific hypothesis to test]
- [ ] [Question to ask in next interview]

## Recommended Actions
[Only if supported by 3+ sources. Otherwise: "No recommendations; need more data."]

## Sources Used
- [file paths]

## Claims Ledger
| Claim | Type | Source |
|-------|------|--------|
| ... | Evidence | [file:line] |

Step 5: Copy to History & Update Tracker

• Copy to history/synthesizing-voc/voc-synthesis-YYYY-MM-DD.md
• Update alerts/stale-outputs.md

Quick Reference

Item	Value
Minimum sources	3
Quote format	Verbatim with source attribution
PII handling	Redact as [REDACTED]
Unknown segments	Label "Unknown", never infer

Common Mistakes

• Paraphrasing: "User was frustrated" → Use exact words: "This is so annoying"
• Inferring sentiment: "They seemed happy" → Only state what's written
• Assuming segments: "Probably enterprise" → Only use explicit segment data
• Recommending without evidence: "We should build X" → Need 3+ sources
• Counting wrong: "Many users said" → Exact count: "4 of 7 sources"

Verification Checklist

• Counted sources (minimum 3)
• All quotes are verbatim (no paraphrasing)
• Every quote has source file attribution
• Segment/Impact marked "Unknown" if not explicit
• PII redacted as [REDACTED]
• Frequency is exact count, not "many" or "some"
• Recommendations backed by 3+ sources
• Metadata header complete
• Copied to history, tracker updated

Evidence Tracking

Claim	Type	Source
[Theme exists]	Evidence	[3+ files cited]
[User segment]	Evidence/Unknown	[explicit in source or "Unknown"]
[Recommendation]	Evidence	[3+ sources supporting]