← Back to Index

Customer Feedback Intelligence System

Created: Feb 17, 2026
Purpose: Automated feedback extraction → structured data → smart routing

Problem Statement

Current state: Customer feedback scattered across emails, support tickets, meeting notes. Manual triage required. Signals get lost.

Desired state: Automated detection, classification, routing. Council sees ops feedback weekly. Dev team sees technical issues pre-meeting. Patterns surface automatically.

System Architecture

1. Input Sources

2. Feedback Detection Layer

LLM-based classifier scans for:

Filters out: Orders, confirmations, spam, automated notifications

3. Classification & Enrichment

Type:

Urgency (1-5):

Frequency tracking:

Customer context:

4. Storage Options

Option A: Zoho CRM Custom Module (Recommended long-term)

Option B: Google Sheet (Quick start)

Option C: Hybrid (Best of both)

5. Routing & Notifications

Ops feedback:

Technical feedback:

Product feedback:

Data Schema

Feedback Entry:
{
  "id": "unique_id",
  "date": "2026-02-17T10:30:00Z",
  "customer": {
    "name": "Millstone K12",
    "email": "bmcnamara@millstone.k12.nj.us",
    "segment": "K-12 District",
    "deal_value": 15000
  },
  "source": {
    "type": "Email",
    "id": "gmail:18d9a1b2c3d4e5f6",
    "thread_subject": "ZTAG Training Feedback"
  },
  "classification": {
    "type": "Operational",
    "category": "Feature Request",
    "urgency": 3,
    "sentiment": "Positive"
  },
  "content": {
    "summary": "School requests digital lesson plans and wristbands for students",
    "raw_text": "[full email quote]",
    "keywords": ["lesson plans", "wristbands", "students"]
  },
  "tracking": {
    "frequency": 1,
    "similar_count": 7,
    "first_seen": "2026-02-17",
    "last_seen": "2026-02-17"
  },
  "routing": {
    "assigned_to": "Council",
    "status": "New",
    "related_system": "Training",
    "priority_score": 6.5
  },
  "resolution": {
    "notes": "",
    "resolved_date": null,
    "outcome": ""
  }
}

Implementation Plan

Phase 1: Shadow Mode (Week 1-2)

Goal: Prove detection accuracy

Build:

  1. Email/ticket scanner (feedback vs noise)
  2. Classification engine (ops vs technical)
  3. Urgency/frequency algorithms
  4. Log to Google Sheet for review

Test:

Deliverable: Working scanner + 50 validated entries

Timeline: 1 week (aggressive) or 2 weeks (thorough)

Phase 2: Automation (Week 3-4)

Goal: Hands-free operation

Build:

  1. Zoho CRM custom module setup (if chosen)
  2. Auto-routing to Telegram groups
  3. Deduplication logic (merge similar feedback)
  4. Weekly digest formatting
  5. Manual override interface (mark as "not feedback")

Test:

Deliverable: Fully automated feedback flow

Timeline: 1-2 weeks

Phase 3: Intelligence Layer (Month 2)

Goal: Surface insights, not just data

Build:

  1. Pattern detection ("3+ similar = trend")
  2. Customer segment analysis (K-12 vs mobile operators)
  3. Impact scoring (high-value customer = boost urgency)
  4. Pre-meeting briefings (auto-generate agendas)
  5. Trend reports ("Battery complaints up 40% this month")

Test:

Deliverable: Proactive intelligence, not reactive logging

Timeline: 2-4 weeks

Example Use Cases

Use Case 1: Bug Pattern Detection

Input: 3 emails from different districts mention "wristband won't turn on"

System:

  1. Detects feedback type: Technical / Bug
  2. Frequency count: 3 occurrences in 2 weeks
  3. Urgency: 4 (multiple customers affected)
  4. Routes to Development group
  5. Links to Code 5 power management component
  6. Summary: "Wristband power-on failure reported by 3 districts (2 weeks)"

Action: Dev team investigates before next UTF Labs meeting

Use Case 2: Training Process Improvement

Input: 5 schools mention "lesson plans would help trainers"

System:

  1. Detects feedback type: Operational / Feature Request
  2. Frequency count: 5 occurrences in 3 weeks
  3. Urgency: 3 (nice-to-have, no blocker)
  4. Routes to Operations & Process group
  5. Summary: "5 districts requested digital lesson plans for trainers"

Action: Council discusses whether to create training materials (ops decision)

Use Case 3: Positive Validation

Input: School emails "teachers love the engagement data dashboard"

System:

  1. Detects feedback type: Product / Validation
  2. Sentiment: Positive
  3. Urgency: 1 (FYI, keep doing this)
  4. Routes to Strategy & Vision group
  5. Summary: "Positive feedback: engagement dashboard well-received"

Action: Reinforce this feature in marketing/training

Questions for Quan

  1. Storage preference:

    • Zoho CRM custom module (long-term best)
    • Google Sheet (quick start)
    • Hybrid (sheet now, CRM later)

  2. Review cadence:

    • Weekly digest only
    • Real-time high-urgency alerts (urgency 4-5)
    • Both (digest + urgent alerts)

  3. Input scope:

    • Email + Zoho Desk only
    • Also scan Fathom meeting transcripts for feedback
    • Start narrow, expand later

  4. Phase 1 timeline:

    • 1 week (aggressive, shadow mode ASAP)
    • 2 weeks (thorough, more testing)

  5. Integration:

    • Feed into escape velocity metrics (operational independence)
    • Standalone system for now

  6. Zoho Desk access:

    • Already have API credentials
    • Need to set up (like Zoho CRM OAuth)

Technical Notes

Detection algorithm:

Deduplication:

Privacy:

Performance:

Success Metrics

Phase 1 (Accuracy):

Phase 2 (Adoption):

Phase 3 (Impact):

Next Steps

  1. Quan: Answer 6 questions above
  2. Minnie: Build Phase 1 scanner (1-2 weeks based on timeline choice)
  3. Quan: Validate 50 sample feedback entries
  4. Minnie: Tune classifier, launch Phase 2 automation
  5. Team: Start using feedback intelligence in Council + Dev meetings

This is escape velocity in action: AI detects signals → routes to right people → decisions informed by data → Quan freed from manual triage.

Let's build it. 🎯