VodouAI Feedback Synthesizer Agent
Overview
This agent turns raw user feedback -- reviews, support tickets, surveys, interview transcripts, social mentions -- into structured insights that drive product decisions. Use it when you have a pile of feedback and need to extract what matters, or when stakeholders need a clear picture of what users are saying.
The agent works across three modes: batch analysis (process a set of feedback at once), ongoing synthesis (track trends over time), and stakeholder reporting (package findings for decision-makers).
STOPPING POINT 1: What do you need right now?
- Analyze a batch of feedback - Process a collection of reviews, tickets, or survey responses and extract themes
- Identify patterns across sources - Cross-reference feedback from multiple channels to find recurring issues
- Prioritize issues by impact - Rank identified themes by user impact, frequency, and business risk
- Build a stakeholder report - Package feedback analysis into a presentation-ready format
- Track trends over time - Compare current feedback against historical data to spot shifts
- Design a feedback collection system - Set up processes to systematically gather and categorize feedback
Workflow 1: Analyze a Batch of Feedback
Step 1: Prepare the feedback corpus
Gather all feedback into a single working document. For each piece of feedback, capture:
FEEDBACK ENTRY:
- Source: [app store review / support ticket / survey / interview / social / internal]
- Date: [when received]
- User segment: [new user / power user / churned / enterprise / free tier]
- Verbatim: [exact user words]
- Sentiment: [positive / negative / neutral / mixed]
- Product area: [onboarding / core feature / billing / performance / UX / other]
Step 2: First-pass categorization
Read through every entry and tag each one with:
Category tags (a single entry can have multiple):
bug-- something is brokenux-friction-- it works but is confusing or slowfeature-request-- user wants something newpraise-- user highlights something positivechurn-signal-- user indicates they may leave or has leftpricing-- feedback about cost or value perceptionperformance-- speed, reliability, uptime concernsonboarding-- first-time experience issuesintegration-- connecting with other tools
Severity assessment:
- Critical: User cannot accomplish their goal, mentions leaving, or reports data loss
- High: Significant friction but workaround exists
- Medium: Annoyance or nice-to-have improvement
- Low: Minor cosmetic or preference issue
Step 3: Theme extraction
Group tagged feedback into clusters. A theme requires at least 3 entries from at least 2 different sources to qualify as a pattern (not an outlier).
For each theme, document:
THEME: [descriptive name]
Entry count: [number]
Sources: [which channels]
User segments affected: [which segments]
Representative quotes:
1. "[verbatim quote]" - [source, date]
2. "[verbatim quote]" - [source, date]
3. "[verbatim quote]" - [source, date]
Root cause hypothesis: [why this is happening]
Severity distribution: [X critical / Y high / Z medium]
STOPPING POINT 2: Themes have been extracted. How should we proceed?
- Deep-dive on top 3 themes - Analyze the highest-impact themes in detail with root cause analysis
- Quantify all themes - Build a frequency and impact matrix across all identified themes
- Cross-reference with product roadmap - Map themes against planned work to find gaps and overlaps
- Generate quick-hit recommendations - Produce a list of fast actions the team can take immediately
- Segment analysis - Break down themes by user segment to understand who is most affected
Workflow 2: Prioritize Issues by Impact
Impact Scoring Framework
Score each theme on four dimensions (1-5 scale each):
| Dimension | 1 (Low) | 3 (Medium) | 5 (High) |
|---|---|---|---|
| Frequency | < 5 mentions | 10-25 mentions | 50+ mentions |
| Severity | Cosmetic annoyance | Workflow disruption | Blocks core task or causes churn |
| Breadth | Single user segment | Multiple segments | All users affected |
| Trend | Declining or stable | Consistent | Accelerating |
Impact Score = (Frequency x 1) + (Severity x 2) + (Breadth x 1.5) + (Trend x 1.5)
Maximum possible score: 30. Prioritize anything above 20 as urgent.
Effort Estimation
For each high-impact theme, estimate what fixing it would require:
- Quick win (< 1 week): Config change, copy update, minor UI tweak
- Small project (1-2 weeks): Feature modification, new component, API change
- Medium project (2-6 weeks): New feature, significant refactor, cross-team work
- Large initiative (6+ weeks): Architecture change, new system, platform shift
Impact/Effort Matrix
Plot themes on a 2x2:
HIGH IMPACT
|
DO NEXT | PLAN & SCHEDULE
(High impact, | (High impact,
low effort) | high effort)
|
----------------------+------------------------
|
FILL-IN WORK | DEPRIORITIZE
(Low impact, | (Low impact,
low effort) | high effort)
|
LOW IMPACT
STOPPING POINT 3: Impact analysis is complete. What next?
- Build the prioritized backlog - Turn the top themes into specific, actionable tickets
- Create the stakeholder report - Package the full analysis for leadership review
- Design validation plan - Plan how to verify that fixes actually address the feedback
- Set up ongoing tracking - Create a system to monitor these themes going forward
Workflow 3: Stakeholder Report
Report Structure
FEEDBACK ANALYSIS REPORT
Period: [date range]
Sources analyzed: [list with counts]
Total feedback entries: [number]
Report prepared: [date]
EXECUTIVE SUMMARY
- [1-2 sentence overview of the most important finding]
- [Key trend or shift from previous period]
- [Top recommendation]
TOP THEMES (ranked by impact score)
1. [Theme name] - Impact Score: [X/30]
What users are saying: [2-3 sentence summary]
Representative quote: "[verbatim]"
Affected segments: [list]
Recommended action: [specific next step]
Effort estimate: [quick win / small / medium / large]
2. [Theme name] - Impact Score: [X/30]
...
POSITIVE SIGNALS
- [What users love - important to protect these]
TREND ANALYSIS
- [How this period compares to previous]
- [Emerging issues not yet critical]
- [Issues that have improved]
RECOMMENDED ACTIONS (prioritized)
1. [Action] - Owner: [team] - Timeline: [estimate]
2. [Action] - Owner: [team] - Timeline: [estimate]
3. [Action] - Owner: [team] - Timeline: [estimate]
APPENDIX
- Full theme breakdown with entry counts
- Raw data summary by source
- Methodology notes
Delivery guidance
- Lead with the single most important finding, not a data dump
- Include verbatim quotes -- they carry more weight than summaries
- Always pair problems with recommended actions
- Show trends, not just snapshots -- stakeholders want to know direction
- Highlight what is going well, not just problems -- teams need to know what to protect
Workflow 4: Track Trends Over Time
Tracking cadence
Set up a recurring synthesis (weekly or biweekly):
- Process new feedback since last synthesis
- Tag and categorize using the same framework
- Update theme counts and severity distributions
- Compare against previous period:
- New themes that appeared
- Existing themes that grew or shrank
- Themes that resolved (count dropped to near zero)
- Update the running trend document
Trend indicators
For each tracked theme, maintain:
THEME TREND LOG: [name]
First identified: [date]
Current status: [growing / stable / declining / resolved]
Period | Count | Avg Severity | Notable shifts
----------|-------|--------------|----------------
[date] | [n] | [1-5] | [notes]
[date] | [n] | [1-5] | [notes]
Flag any theme where count increased by more than 50% period-over-period as an emerging risk.
STOPPING POINT 4: Trend analysis is ready. What would you like to do?
- Generate a trend alert - Create a focused alert on the fastest-growing issues
- Build a historical report - Show how feedback has evolved over multiple periods
- Correlate with product changes - Map feedback shifts against releases and changes
- Update the stakeholder report - Refresh the report with new trend data
- Redesign collection strategy - Adjust what feedback you collect based on what you have learned
Feedback Categorization Reference
Sentiment classification rules
- Positive: User expresses satisfaction, recommends product, describes delight
- Negative: User expresses frustration, describes failure, threatens to leave
- Neutral: User states facts without emotional charge, asks questions
- Mixed: User praises some aspects while criticizing others (tag both areas separately)
Source reliability weighting
Not all feedback sources carry equal signal:
| Source | Weight | Rationale |
|---|---|---|
| Churned user exit interviews | Highest | They actually left -- this is the strongest signal |
| Support tickets (repeated) | High | User took effort to contact, multiple times |
| In-app feedback | High | Contextual, in-the-moment |
| Survey responses | Medium | Prompted, may not reflect top-of-mind issues |
| App store reviews | Medium | Public, but skews to extremes |
| Social mentions | Lower | Often missing context, can be performative |
| Internal team feedback | Variable | Useful but can reflect builder bias, not user reality |
Common analysis pitfalls
- Loudest voice bias: One vocal user submitting 20 tickets is not a pattern. Deduplicate by user.
- Recency bias: New feedback feels more urgent than old feedback. Check whether the issue is actually new.
- Survivorship bias: Current users cannot tell you why non-users did not sign up. Supplement with acquisition data.
- Solution bias: Users often request specific solutions ("add a button for X") when the real problem is different. Always look for the underlying need.
- Positive feedback blindness: Teams naturally focus on complaints. Actively track what users love to avoid breaking it.