Social commerce is not a chatbot pasted onto a website. It is commerce-native conversations: every message maps to orders, business units, and associates in your commerce backend.
Architecture that scales
- Messaging gateway — WhatsApp Business API (or simulator for demos) with verified webhooks
- Identity layer — phone → commercetools Associate → Business Unit scope
- Intent router — keywords first, LangGraph agents for free-text with tool calls
- CRM inbox — human agents with order context, proposals, audit trail
- Headless storefronts — same commercetools catalog powers D2C and B2B
Why LangGraph on top of commerce APIs
Free-text distributor messages ("where is order RC-10042?" / "approve the change") need stateful reasoning:
- Plan which CT API to call
- Execute with BU-scoped credentials
- Format reply for messaging UI
- Escalate to CRM on low confidence
LangGraph makes this debuggable — each node is testable, unlike one-shot prompts.
Social commerce metrics that matter
- Time-to-status for order inquiries
- Proposal approval rate via messaging
- CRM handle time with linked order context
- D2C conversion unaffected by B2B channel load
Takeaway
Start with structured intents and a shared inbox. Add LangChain/LangGraph when free-text volume justifies it. Keep humans in the loop for money-moving actions.
