Skip to main content

S4. Persona Matching + Clustering (AMWAY)

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

1. URL & Personas

  • /personas · P2 (Consumer Insights)

2. Five Lifestyle Personas

PersonaCore Signals
Health InfluencerNutrilite subscription · Instagram posts · share of Diamond ABOs
Mom-of-kidsHome · kids' vitamins · weekday daytime activity
Single HouseholdSmall portions · convenience · late-night activity
SeniorNutritional supplements · phone orders
TrendsetterArtistry new arrivals · SNS campaign responsiveness

3. Data Mix

  • RFM features (combined Customer + ABO)
  • Category affinity (Nutrition / Beauty / Home)
  • Channel share (store / ABO direct / catalog)
  • Social personas (Instagram · Reddit · X keywords)

4. Processing Pipeline (KMeans)

  1. Feature extraction + standardization
  2. KMeans 6 (or elbow)
  3. Per-cluster averages + social enrichment → Sonnet labeling
  4. Labels: "Health Influencer," "Vegan Single Household," "Weekend Family Shopper," etc.

5. Output UI

  • 5 persona cards (distribution bars)
  • 6 cluster cards (label, member count, top 5 representative SKUs)
  • PCA scatter plot

6. Guardrails

  • Do not expose persona labels to the labeled members themselves
  • Prohibit pregnancy / illness inference
  • Flag samples with fewer than 30 members

7. Demo Scenarios

  1. Single-item match → "First-rank Health Influencer 0.62"
  2. Batch of 50,000 members → six clusters auto-labeled
  3. "Campaign to this cluster" → handoff to S5