S2. Persona Chatbot Across 5 Departments (AMWAY)
Published 2026-05-14Updated 2026-06-303 min read
1. URL & Personas
/chat· All personas
2. User Stories
- P1 (ABO Field): "Top 10 ABOs who reached Diamond this quarter and their Downline activity"
- P5 (Compliance): "Protein-campaign ad review + minor block counts"
3. UI
- Left: chat / Right: SSE tool-call trace
- Persona switcher + four suggested queries
4. 12 Tools
| Tool | Description |
|---|---|
semantic_search | OpenSearch BM25 + KNN + Rerank (multilingual) |
neptune_query | openCypher |
abo_tree_query | ABO tree depth query (S9-A) |
subscription_lifecycle | Subscription state · churn signature (S10-A) |
campaign_simulator | Bayesian ROAS |
behavior_change_detect | Statistical signature |
social_trend_join | Social multilingual |
weather_join | KMA · global weather |
economic_join | FX · prices (multi-country) |
regulation_join | Direct-selling regulations |
cluster_run | KMeans + LLM labeling |
pii_mask | PII masking |
5. Per-persona Variations
| P | system prompt keywords | Primary tools |
|---|---|---|
| P1 | ABO · Sponsor · Downline · PV/BV | abo_tree_query, campaign_simulator |
| P2 | Cohort · LTV · SOV | cluster_run, social_trend_join |
| P3 | Accuracy · drift | behavior_change_detect, neptune_query |
| P4 | Level · NPS · subscription | subscription_lifecycle, semantic_search |
| P5 | Regulation · consent · minors | regulation_join, pii_mask |
6. Guardrails
- Input: 4 topics + direct selling + minors + HFF
- Output: PII masking; citations are mandatory for regulatory answers
- Tools: write tools are restricted to P3 and P5
7. Demo Scenarios
- (P1) "Top 10 ABOs reaching Diamond this quarter" → abo_tree_query + neptune_query
- (P1 → P4 switcher) Same query → tone shifts (level / NPS lens)
- (P5) "Review Nutrilite advertising" → regulation_join + list of high-risk SKUs