S9-Mo. Live-Commerce Broadcast Attribution (Momo-Specific)
Published 2026-05-14Updated 2026-06-302 min read
Momo differentiation point ⭐ — Live-broadcast viewing, pin, and purchase events are ingested in real time via Kinesis for per-host, per-SKU, and per-time-slot attribution.
1. URL · Persona
/live-attribution· P1 (Marketing)
2. User Story
P1 — Yesterday's 19:00 live broadcast (1 hour) — 50K viewers, 8K carts, 3K purchases, 120M NTD GMV. Per-host, per-SKU, and per-time-slot effect on a single screen.
3. Data Mix
| Data | Source |
|---|---|
| LiveStream metadata | Neptune |
| LiveViewer viewing events | Kinesis → Neptune |
| LiveStreamPurchase | Neptune (~50K) |
| Pin events (host pinning an SKU) | LiveStream attribute (time sequence) |
| External SNS reactions | SocialSignal |
4. Processing Pipeline
1. Select a single live broadcast
2. Compute the viewer → cart → purchase funnel
3. Aggregate per-host GMV and conversion rate
4. Attribute purchases within ±5 minutes of an SKU pin
5. Viewing time vs purchase probability curve (survival analysis)
5. Output UI
- Left: Funnel chart (Viewing → Cart → Purchase)
- Center: Per-host GMV bars + ratings
- Right: SKU pin vs purchase timeline (interactive)
- Bottom: Viewing-time distribution + purchase probability
6. Guardrails
- PII masking (viewers)
- Host business-information protection
7. Demo Scenarios
- Yesterday's 19:00 live broadcast → 50K viewers → 8K carts → 3,200 purchases → 120M NTD GMV
- Host A vs B → Same SKU pin, conversion 12% vs 4%
- 5-minute GMV spike after SKU pin → automatic attribution