Skip to content

README.md

Latest commit

 

History

History
115 lines (87 loc) · 4.62 KB

README.md

File metadata and controls

115 lines (87 loc) · 4.62 KB

ozi-insights

Customer insight layer for the Ozi app. This workspace holds synthetic support data grounded in real Play Store signal, to be used as input for a structured analysis pass.

What's Here

apps/ozi-insights/
  data/
    freshdesk-synthetic.json    ← 30 synthetic Freshdesk tickets (raw data only)
  README.md

Why This Exists

Ozi's Play Store reviews (3.9★, 85 reviews, 18×1★) surfaced 8 recurring pain point categories. Rather than jumping straight to solutions, this dataset simulates what Ozi's Freshdesk inbox would look like — realistic customer support transcripts grounded in those pain points.

Design principle: The data is raw. No categories are injected into the conversations. A future analysis pass reads this file cold and must discover patterns programmatically, the same way a real support analyst would.

Ticket Schema

Each object in freshdesk-synthetic.json:

{
  "id": "TKT-001",
  "subject": "...",
  "status": "open | pending | resolved | closed",
  "priority": "low | medium | high | urgent",
  "created_at": "ISO 8601 with +05:30 offset",
  "updated_at": "ISO 8601 with +05:30 offset",
  "resolved_at": "ISO 8601 | null",
  "resolution_time_minutes": 136,
  "csat_score": 1,
  "tags": ["delivery-delay", "sla-breach"],
  "category": "delivery-delay",
  "channel": "chat | email | phone",
  "requester": {
    "name": "Priya Sharma",
    "location": "Gurgaon | Noida | Delhi | Mumbai"
  },
  "conversation": [
    {
      "role": "customer | agent",
      "body": "...",
      "timestamp": "ISO 8601 with +05:30 offset"
    }
  ]
}

Field Notes

Field Notes
status open = active/unresolved. pending = waiting on customer/internal. resolved = closed with action. closed = auto-closed or no-action
resolved_at null for ghost-support tickets and unresolved cases
resolution_time_minutes null when unresolved. Values >1440 indicate next-day+ resolution
csat_score 1–5. null when not rated (open tickets, or customer did not rate)
tags Multi-label. One ticket can have delivery-delay + support-ghost together
category Single primary category — the dominant pain point

Pain Point Categories

Category Tag Description Ticket Count
delivery-delay Order arrived significantly later than the 30-min SLA 10
support-ghost Agent went unresponsive mid-conversation or ticket closed without resolution 7
wrong-product Incorrect, defective, used, or misrepresented product delivered 5
no-cancellation Customer wanted to cancel but found no self-serve option in the app 3
serviceability Customer in unsupported city (Delhi, Mumbai) misled by app UX 2
pricing Product perceived as too expensive vs offline alternatives 2
app-bug Feature crash (demo booking) with cascading ops failure 1

Realism Signals to Note

  • Hinglish in conversations: Tickets 026, 029, and others include Hindi-English mix, authentic for Delhi NCR parent demographic
  • Ghost-support pattern: TKT-004, 012, 013, 025 — agent responds then conversation ends with no follow-up, status remains open
  • Escalating frustration: TKT-003, 010, 015 — customer tone degrades across messages as resolution stalls
  • Broken promises: TKT-005 (overnight delivery + no sorry coupon), TKT-030 (no-show after manual booking) — compound failures
  • CSAT distribution: Not all 1s. Scores range 1–4 based on resolution quality. Some tickets have null CSAT (unresolved or not rated)

What the Analysis Pass Should Compute

When this dataset is fed into an analysis script:

  1. Frequency by category — which issue type appears most?
  2. Average CSAT by category — which issue creates most dissatisfaction?
  3. Resolution rate by category — % of tickets that are resolved | closed vs open | pending
  4. Average resolution time by category — where is the operational lag worst?
  5. Co-occurrence of tags — e.g. delivery-delay + support-ghost together signals a compounding failure mode
  6. Ghost rate — % of tickets with resolved_at: null and last message from customer

Each category that scores high across multiple metrics becomes an experiment candidate, which then maps to an MVP spec for engineers.

Source Signal

App: OZi: Fast. Trusted. For Kids. (com.ozi.user) Reviews scraped: March 21, 2026 Rating at time of scrape: 3.9★ overall, 85 reviews — 57×5★ / 5×4★ / 3×3★ / 2×2★ / 18×1★ Coverage: Gurgaon and Noida only (as of March 2026)