Product Feedback Synthesizer

Workflow

Step 1 — Ingest & Inventory

1. Identify all available feedback sources (files, URLs, pasted text).
2. For each source, record: source name, channel type (survey / review / ticket / interview / social), date range, item count.
3. Produce a Source Inventory Table before proceeding:

| Source            | Channel   | Date Range      | Items |
|-------------------|-----------|-----------------|-------|
| NPS survey export | Survey    | Jan–Mar 2024    |  412  |
| App Store reviews | Review    | Q1 2024         |   87  |
| Support tickets   | Ticket    | Feb 2024        |  230  |

Validation checkpoint: Warn if total items < 10 (limited statistical reliability). Flag inconsistent date ranges across sources before synthesis.

Step 2 — Clean & Normalise

1. Remove exact duplicates (same text, same timestamp).
2. Strip PII (names, emails, account IDs → [REDACTED]).
3. Standardise sentiment labels: Positive / Neutral / Negative.
4. Assign each item a unique ID: SRC-### (e.g., NPS-001, TKT-047).

Validation checkpoint: If > 15% of items are duplicates, note this in the summary — it may indicate a data export issue.

Step 3 — Categorise & Tag

For each feedback item assign:

• Primary Theme: Usability · Performance · Pricing · Feature Request · Bug Report · Onboarding · Support · Other
• Sentiment Score: Positive (+1) / Neutral (0) / Negative (−1)
• Priority Signal: High / Medium / Low — language like "can't use", "losing customers", or "deal-breaker" = High

Produce a Categorisation Sample (first 5–10 items) to confirm taxonomy before bulk analysis:

| ID      | Source  | Theme           | Sentiment | Priority | Verbatim (truncated)              |
|---------|---------|-----------------|-----------|----------|-----------------------------------|
| NPS-001 | Survey  | Feature Request | Positive  | Medium   | "Would love a dark mode..."       |
| TKT-047 | Ticket  | Bug Report      | Negative  | High     | "Export crashes every time I..."  |
| REV-012 | Review  | Performance     | Negative  | High     | "App lags badly on Android..."    |

Validation checkpoint: For large (> 100 items) or domain-specific datasets, ask the user to confirm or adjust the theme taxonomy before proceeding.

Step 4 — Quantify & Rank

Calculate per theme:

• Volume: count and % of total
• Sentiment Ratio: % Positive, % Neutral, % Negative
• Weighted Priority Score = (High×3 + Medium×2 + Low×1) / item count

Produce a Priority Matrix:

| Rank | Theme           | Volume | % Total | Sentiment Ratio (+/0/−) | Priority Score |
|------|-----------------|--------|---------|-------------------------|----------------|
|  1   | Performance     |   148  |  28.5%  | 5% / 10% / 85%          |      2.74      |
|  2   | Feature Request |   112  |  21.6%  | 70% / 20% / 10%         |      1.88      |
|  3   | Bug Report      |    89  |  17.1%  | 0% / 5% / 95%           |      2.91      |

Step 5 — Synthesise Insights

For each top-ranked theme (up to 5), write a structured insight block:

## Insight: [Theme Name]

**Signal strength:** [e.g., 148 mentions, 85% negative sentiment, Priority Score 2.74]

**What users say (representative verbatims):**
- "The app freezes for 10–15 seconds after uploading files." (REV-023)
- "Performance has gotten worse with every update." (NPS-187)
- "Unusable on mobile — takes 30 seconds to load a dashboard." (TKT-091)

**Pattern summary:**
[2–3 sentences describing the core user experience problem, affected segments, and when it occurs.]

**Recommended action:**
[1–2 concrete, specific actions — e.g., "Profile and optimise dashboard load time on Android; target < 3s. Prioritise before next feature release."]

**Business impact estimate:**
[Link to a metric where possible — e.g., "Appears in 40% of 1-star reviews; addressing may improve App Store rating and reduce churn in mobile segment."]

Step 6 — Deliver Report

Assemble the final output in the following sections (see REPORT_TEMPLATE.md for the full ready-to-fill template):

1. Report header — period, total items analysed, sources list.
2. Executive Summary — 3–5 sentences: top finding, most urgent issue, headline recommendation.
3. Priority Matrix — table from Step 4.
4. Top Insights — insight blocks from Step 5, ranked by Priority Score.
5. Feature Request Ranking — table sorted by mention frequency, with mention count, % positive sentiment, and a sample quote per entry.
6. Voice of Customer Highlights — 5–8 high-impact verbatim quotes, one per major theme, with source IDs.
7. Recommended Next Steps — one immediate action (within 1 sprint), one short-term action (within 1 quarter), one strategic roadmap consideration.
8. Data Quality Notes — caveats on sample size, date gaps, duplicate rate, or missing sources.

For a pre-formatted, copy-paste-ready version of this structure, refer to REPORT_TEMPLATE.md.

Worked Example

See EXAMPLES.md for a complete end-to-end walkthrough using 3 app-store reviews and 2 support ticket summaries, including Step 1 source inventory, Step 3 categorisation sample, and a full Step 5 insight block.

Processing Guidelines

• Tone: Report findings factually. Do not soften negative feedback or editorialize.
• Verbatims: Always include source IDs so findings are traceable.
• Uncertainty: Label ambiguous sentiment (sarcasm, mixed signals) as Neutral and note the ambiguity.
• Scope: If the user requests competitive analysis or persona development beyond the provided data, flag that additional data collection is needed.
• Iterative mode: When feedback arrives in batches, re-run Steps 3–6 on the combined dataset rather than merging reports post-hoc.