Skip to main content

S5. Omnichannel Campaign ROAS (AMWAY)

Published 2026-05-14Updated 2026-06-302 min read

1. URL & Personas

  • /campaign-roas · P1 (ABO Field)

2. Candidate Channels

  • SMS · email · KakaoTalk/Line · Instagram ads · in-house store banners · ABO direct sale (Sponsor) · catalog · influencer

3. Data Mix

DataSource
Historical campaignsSnowflake CAMPAIGN_MART
Channel sell-throughOrderTransaction + ABODirectSale
SNS responseSocialSignal (Instagram · Reddit · X)
Search trendsMultilingual search trends

4. Processing Pipeline

  1. Extract channel · country priors from Snowflake
  2. 1,000 MCMC samples (PyMC) → posterior
  3. ROAS distribution for the channel mix
  4. Attribution (last-touch / linear / time-decay)

5. Output UI

  • Recommended channel mix (donut)
  • ROAS distribution (violin)
  • Marginal efficiency by channel and country
  • Separately display trend vs. SNS impact

6. Guardrails

  • Automatically exclude consent-missing members
  • Automatically exclude minor-targeted campaigns
  • Block direct-selling-regulation-violating text

7. Demo Scenarios

  1. Budget KRW 100M / Nutrilite new product / 50K Gold ABOs → recommended mix
  2. Raise ABO direct-sale share by 20% → instant ROAS shift
  3. "Trend effect vs. campaign effect" decomposed