Appearance
VoC Data Sources & Feedback Taxonomy
Categorizing the 15+ feedback channels a VoC agent should ingest, and how to prioritize them.
Three Feedback Categories
1. Direct Feedback (Explicitly Given)
Customer intentionally provides feedback:
| Source | Signal Type | Richness | Volume |
|---|---|---|---|
| NPS surveys | Quantitative + verbatim | Medium | High |
| CSAT surveys | Satisfaction score | Low | High |
| In-app feedback widgets | Feature requests, bugs | Medium | Medium |
| Feedback portals (Canny, Productboard) | Structured feature requests | High | Low |
| QBR meetings | Strategic priorities | Very High | Very Low |
| Advisory board sessions | Roadmap input | Very High | Very Low |
2. Indirect Feedback (Observed)
Customer behavior/words captured during normal interactions:
| Source | Signal Type | Richness | Volume |
|---|---|---|---|
| Gong call transcripts | Objections, requests, sentiment | Very High | High |
| Zendesk/Intercom tickets | Pain points, workarounds | High | High |
| Email threads (Gmail, Outlook) | Requests, escalations | Medium | High |
| Slack/Teams messages | Real-time reactions | Medium | Very High |
| Social media mentions | Brand sentiment, complaints | Low | Medium |
| G2/Capterra reviews | Competitive comparison | Medium | Low |
| Community forums | Power user needs | High | Low |
3. Inferred Feedback (Behavioral)
What customers do (not say):
| Source | Signal Type | Richness | Volume |
|---|---|---|---|
| Product usage analytics | Feature adoption, drop-off | High | Very High |
| Churn/renewal data | Retention signals | Medium | Medium |
| Support ticket frequency | Friction patterns | Medium | High |
| Login frequency/recency | Engagement health | Low | Very High |
| Feature usage heatmaps | UI/UX pain points | Medium | High |
| Expansion/contraction | Value realization | Medium | Medium |
Prioritization for Building a VoC Agent
Tier 1 — Start Here
- Gong transcripts — highest signal-to-noise ratio, richest unstructured data
- Support tickets (Zendesk/Intercom) — direct pain points with volume
- Salesforce CRM — deal context, win/loss reasons, account health
Tier 2 — Add Next
- Product usage analytics — behavioral signals that validate verbal feedback
- NPS/CSAT surveys — quantitative baseline with verbatims
- Email threads — escalation patterns, executive sentiment
Tier 3 — Full Coverage
- Slack/Teams — real-time customer reactions (if customer-facing channels exist)
- Review sites (G2, Capterra) — competitive positioning
- Community forums — power user feature requests
- Social media — brand health monitoring
Cross-Channel Correlation
The real power of VoC agents is connecting signals across sources:
- Issue appears in G2 reviews -> escalates through support tickets -> surfaces in sales calls -> impacts social sentiment
- Siloed analysis per channel misses the trajectory
- Build the agent to trace themes across all sources, not just aggregate per-source
Custom Taxonomy
Following Enterpret's approach:
- Train the model on your product's specific language and terminology
- Generic sentiment models miss domain-specific nuance
- Example: "the workflow is clunky" means something different in expense management vs. CRM
- Per-customer NLP models outperform one-size-fits-all
Key Takeaways
- Three categories: Direct (surveys), Indirect (calls/tickets), Inferred (usage data) — a complete VoC agent needs all three
- Start with Gong + Support tickets + Salesforce (Tier 1) before expanding
- Cross-channel correlation is the killer feature — tracing a theme from review to ticket to call to churn
- Custom taxonomy trained on your product language dramatically improves signal quality
- Inferred signals (behavioral data) validate or contradict what customers say — always cross-reference