Pa feedback aggregator

Feedback Aggregator

Collects product feedback from multiple sources, clusters it into themes, and produces structured reports for the product team. Part of the Product Adoption & Value Realisation pillar.

This is a collection and analysis skill. It aggregates and themes feedback. It does not prioritise product roadmap decisions (that is product management) or advocate for specific customers (that is the CSM's job using this data as evidence).

When to Run

• Continuous: Scans incoming support tickets and call transcripts for feedback signals
• Scheduled: Monthly aggregation and theming report
• On-demand: When a CSM or product manager requests feedback analysis on a specific topic or feature area

Core Execution Logic

Step 1: Collect Feedback from All Sources

Source	Collection Method	Signal Type	Confidence
Support tickets	Scan for feature request tags, "wish," "would be great if," "need to be able to" language	Direct product feedback	High -- customer explicitly stating a need
Call transcripts	Extract product discussion segments, feature questions, frustration language, workaround descriptions	Conversational feedback	Medium -- context matters, may be speculative
NPS/CSAT surveys	Extract free-text responses mentioning product capabilities	Sentiment-linked feedback	High -- customer volunteered the feedback
CSM notes	Scan CRM notes for product-related observations and customer quotes	Relationship-context feedback	Medium -- filtered through CSM interpretation
Feature request log	Pull from formal feature request system if one exists	Formal requests	High -- explicitly logged
Community forum	Extract posts discussing product capabilities or gaps (from ca-community-monitor if available)	Peer-validated feedback	Medium -- may represent vocal minority

Step 2: Normalise and Deduplicate

For each feedback item captured:

1. Normalise language: "need better reporting" and "reporting is too basic" and "we need custom dashboards" may all be the same theme. Group by intent, not by exact phrasing
2. Deduplicate per account: Multiple tickets from the same account about the same issue count as one feedback item with severity weight, not multiple independent requests. This prevents a single noisy customer from dominating the analysis
3. Tag metadata: For each item, record:
   • Account ID, ARR, segment
   • Source (ticket/call/survey/note/formal request)
   • Date captured
   • Sentiment (request vs. frustrated complaint vs. blocking issue)
   • Contact role (end user, admin, champion, executive -- executive feedback carries different weight)

Step 3: Cluster into Themes

Group feedback into a two-level taxonomy:

Level 1 -- Category:

Category	Definition
Feature gap	A capability the product does not have that customers want
Feature improvement	An existing feature that needs enhancement
UX/usability	The feature exists but is too difficult or unintuitive to use
Performance	Speed, reliability, scale, uptime issues
Integration	Connections to other tools, API capabilities, data import/export
Documentation	Missing or outdated help content, unclear guidance

Level 2 -- Specific theme within category: Each category contains specific themes (e.g., within "Feature gap": "Custom reporting dashboards," "Advanced permission controls," "Mobile app"). Themes emerge from the clustering -- do not force feedback into predetermined themes.

Step 4: Rank Themes by Impact

For each theme, compute:

Weight Factor	Computation	Why It Matters
Breadth	Number of unique accounts with this feedback	Wide impact vs. single-customer issue
Revenue weight	Sum of ARR across requesting accounts	Financial materiality
Health correlation	% of accounts with this feedback that also have declining health scores	Retention impact
Churn correlation	% of accounts that churned in the last 12 months that had this feedback before churning	Predictive churn signal
Sentiment intensity	% of feedback items tagged as "blocking issue" or "frustrated complaint" vs. "request"	Urgency signal

Composite impact score = (Breadth * 0.25) + (Revenue weight normalised * 0.25) + (Health correlation * 0.20) + (Churn correlation * 0.20) + (Sentiment intensity * 0.10)

Step 5: Generate Reports

Monthly Product Feedback Report: For the product team, CS leadership, and cross-functional review.

{
  "report_period": "February 2026",
  "total_feedback_items": 142,
  "unique_accounts": 38,
  "new_themes_this_month": 3,
  "top_themes": [
    {
      "theme": "Custom reporting dashboards",
      "category": "feature_gap",
      "impact_score": 8.4,
      "accounts": 12,
      "total_arr": 890000,
      "sentiment_breakdown": { "request": 7, "frustrated": 4, "blocking": 1 },
      "health_correlation": "3 of 12 accounts also have declining health scores",
      "churn_correlation": "2 accounts that churned in the last 12 months had this feedback",
      "representative_feedback": "Our analytics team exports data to build reports manually. It takes 4 hours a week. If we could build custom dashboards inside the platform, that time goes away.",
      "roadmap_status": "Planned -- Q3 2026",
      "summary": "12 accounts across mid-market and enterprise requesting customisable reporting. Correlated with adoption plateaus in the analytics feature set. Planned for Q3 -- product team should consider accelerating given the churn correlation."
    }
  ],
  "trending": [
    { "theme": "API rate limit increases", "direction": "up", "new_accounts_this_month": 4 }
  ],
  "resolved_this_month": [
    { "theme": "Bulk import performance", "resolution": "Performance improvement shipped in v3.2", "accounts_affected": 8 }
  ]
}

Per-account feedback view: For CSMs preparing for QBRs or renewal conversations.

{
  "account_id": "string",
  "active_feedback_items": 3,
  "themes": [
    { "theme": "Custom reporting", "status": "planned_q3", "first_raised": "2025-09-15", "times_raised": 4 }
  ],
  "resolved_items": 2
}

Handoff to Human

Decision Point	Context Provided	What the Human Decides
Theme validation	Clustered themes with evidence	Whether the clustering is correct -- did the agent merge things that are distinct, or split things that belong together?
Impact assessment	Composite scores with component breakdown	Whether the weighting reflects the business reality. A theme with low breadth but one account threatening to churn may deserve higher priority than the score suggests
Product advocacy	Ranked themes with evidence and roadmap status	How to present this to the product team. The CSM uses the data to advocate; the data alone does not set priorities
Customer communication	Per-account feedback view with roadmap status	How and when to communicate roadmap updates to customers who raised the feedback

Confidence and Limitations

• High confidence for feedback collection and counting -- aggregating tickets and requests is deterministic
• High confidence for deduplication per account -- same account, same topic, same item
• Medium confidence for theme clustering -- NLP-based grouping may merge distinct issues or split related ones. Human review of the top 10 themes each month is recommended before presenting to the product team
• Medium confidence for impact scoring -- the weighting formula is a sensible default but should be calibrated against actual churn data over time
• Low confidence for churn correlation -- correlation is not causation. A churned account that mentioned a feature gap may have churned for an unrelated reason. Present as "correlated with" not "caused by"
• Feedback from vocal customers may be overrepresented even after deduplication. One account raising 15 tickets on the same issue counts as one feedback item, but the intensity signal may still skew perception. The breadth metric (unique accounts) is the most reliable measure of true demand
• Cannot assess the feasibility or cost of building a requested feature -- that is product management territory
• Qualitative feedback (verbatims from calls and surveys) requires NLP that may miss nuance, sarcasm, or cultural context. High-impact themes should be verified against the raw source material

Dependencies

Required:

• Support platform API (ticket content, tags, metadata)
• CRM API (account data for ARR weighting and contact role identification)

Strongly recommended:

• Call transcript integration (Gong, Chorus) for conversational feedback
• Survey/NPS tool for sentiment-linked feedback
• bi-health-score (for health correlation analysis)
• ic-feature-request-tracker (for roadmap status linkage)

Downstream consumers:

• Product team (primary consumer -- monthly report for roadmap input)
• ic-cross-func-prep (CS-product sync materials)
• cc-qbr-deck-builder (per-account feedback view for QBR context)
• CS leadership (feedback-correlated churn data for strategic planning)

References

• references/feedback-taxonomy.md -- Category definitions, clustering methodology, and deduplication rules