Ecom Conversational Agent Skill

Conversational Commerce

Framework

IRON LAW: Conversation First, Commerce Second

Conversational commerce works because it meets customers WHERE THEY
ALREADY ARE (messaging apps). Forcing a sales pitch into a chat channel
kills trust. The conversation must provide genuine value (answering questions,
solving problems) BEFORE introducing products or purchases.

The sequence: Help → Trust → Recommend → Convert

Platform Comparison (Asia-Pacific Focus)

| Platform | Users (Taiwan) | Commerce Features | Best For |

|----------|---------------|------------------|----------|

| LINE | 21M+ (95% penetration) | LINE Shopping, Rich Menu, LIFF, payment | Taiwan, Japan, Thailand primary channel |

| Instagram DM | ~10M | Shop tags, quick replies, product stickers | Visual products, younger demographic |

| Facebook Messenger | ~18M | Shops integration, automated responses | Broad reach, older demographic |

| WhatsApp | Limited in Taiwan | Catalog, cart, payment (select markets) | SEA, India, Brazil |

| WeChat | ~1M (Taiwan) | Mini Programs, WeChat Pay | China-connected businesses |

Conversation Flow Design

1. Entry Points — How customers start the conversation

• QR code in store, ad click-to-message, website chat widget, social media link

2. Welcome Flow — First 3 messages

• Greet warmly, set expectations, offer navigation options
• Rich menu (LINE) or quick reply buttons — don't ask open-ended questions early

3. Product Discovery — Help them find what they need

• Guided questions: "What are you looking for?" → category → product
• Product cards with image, price, and "Buy" button
• AI-powered recommendations based on conversation context

4. Purchase Flow — Minimize friction

• In-chat checkout where possible (LINE Pay, in-app payment)
• If redirecting to website, deep-link to the specific product (not homepage)
• Order confirmation message with tracking

5. Post-Purchase — Retain and upsell

• Shipping updates via message
• Follow-up: "How's the product?" (7 days after delivery)
• Personalized recommendations based on purchase history

Chatbot vs Human Handoff

| Scenario | Handle with Bot | Hand off to Human |

|----------|----------------|------------------|

| FAQ (hours, shipping, returns) | ✓ | — |

| Product recommendations (simple) | ✓ | — |

| Complex product questions | — | ✓ |

| Complaints / issues | — | ✓ (immediately) |

| High-value purchases | Bot assists → human closes | ✓ |

Key metric: Bot containment rate (% resolved without human) — target 60-70% for mature bots.

Output Format

# Conversational Commerce Plan: {Business}

## Channel Selection
- Primary: {platform} — rationale: {why}
- Entry points: {QR / ad / social / website}

## Conversation Flow
1. Welcome: {message template}
2. Discovery: {question flow}
3. Product Card: {template}
4. Checkout: {in-chat / redirect}
5. Post-purchase: {follow-up sequence}

## Bot vs Human Split
| Scenario | Handler | SLA |
|----------|---------|-----|
| {scenario} | Bot/Human | {response time} |

## KPIs
| Metric | Target |
|--------|--------|
| Response time | < {X} seconds (bot) / < {X} minutes (human) |
| Containment rate | > 60% |
| Conversation-to-purchase rate | > {X%} |
| Customer satisfaction (CSAT) | > 4.0/5 |

Gotchas

• LINE Official Account tiers matter: Free tier limits monthly messages. If you exceed, messages are throttled. Budget for premium tier if customer base > 500.
• Don't over-automate: A bot that can't understand the question and loops "I didn't understand, please choose from the menu" destroys trust faster than no bot at all. Always offer a human escalation path.
• Messaging is asynchronous: Unlike phone calls, customers expect to message and come back later. Design flows that work with interruptions — save cart state, remember context.
• Privacy in messaging: Chat history is personal. Don't share conversations internally without consent, and be transparent about data usage.
• Social commerce is exploding in SEA/Taiwan: LINE Shopping, Instagram Shopping, TikTok Shop — these blur the line between social and commerce. Treat them as primary channels, not add-ons.

References

• For LINE Official Account setup, see references/line-oa-setup.md
• For chatbot NLU design patterns, see references/chatbot-design.md