Stateset iCommerce Architecture: Technical & Business Explainer

Executive summary

Traditional ecommerce forces shoppers through pages, forms, and portals. Stateset compresses the journey to a single conversational thread powered by AI agents, ACP for safe payments, and an event-driven backend that keeps orders, inventory, and accounting in sync. Customers finish in 30-90 seconds; merchants gain 3-5x conversion and seconds-fast support.

Business context

The legacy model expects customers to browse catalogs, fill forms, wait for emails, and open tickets for changes. That drives 60-80% abandonment and slow support. Stateset replaces that with natural language flows, delegated payment, and instant post-purchase help.

AI agent system

Multiple agents collaborate with shared memory and a session manager that routes tool calls through MCP.

CUSTOMER CONVERSATION (ChatGPT)
        |
Stateset AI Agent Layer
  - Commerce Agent (discover, checkout, pay)
  - Customer Service Agent (returns, tracking, warranty)
  - Operations Agent (inventory, routing, sync)
        |
Session Manager & Context Engine
  - Customer memory, KB, RAG, tool routing
        |
Stateset Infrastructure (orders, inventory, payments, fulfillment)

ACP: Agentic Commerce Protocol

ACP standardizes delegated checkout so AI actions stay safe: scoped tokens, explicit consent, HMAC signatures, and idempotent endpoints.

Step 1: POST /checkout_sessions        → create session
Step 2: POST /checkout_sessions/{id}    → buyer, shipping, billing
Step 3: POST /agentic_commerce/delegate_payment → scoped token
Step 4: User confirms in ChatGPT
Step 5: POST /checkout_sessions/{id}/complete → payment + order

MCP: Model Context Protocol

MCP connects ChatGPT to Stateset tools and widgets. 30+ tools cover orders, returns, tracking, and inventory; widgets render carousels, order views, and tracking cards inside ChatGPT.

Retrieval-augmented answers

Agents mix knowledge base articles, customer memory, and real-time data (inventory, order status, pricing) to respond with context and eligibility checks before acting.

Order flow overview

1. Discovery: natural language search, recommendations, cross-sell.
2. Checkout: ACP session creation, shipping capture, delegated payment.
3. Payment: token validation, HMAC, gateway charge, order creation.
4. Routing: sync server selects fulfillment node, reserves inventory.
5. Fulfillment: shipment creation, carrier label, tracking updates.
6. Accounting: double-entry revenue, COGS at ship, refunds when needed.
7. Post-purchase: tracking, returns, exchanges, warranty, proactive issues.

Architecture layers

Data flow (condensed)

Customer intent → AI agent → MCP tool → ACP checkout
   → Sync server routing → Stateset API
   → Orders, inventory, payments, accounting
   → Events → webhooks, email, widgets → customer

Reliability and safety

• Event-driven outbox for reliable delivery and retries.
• Scoped delegated payment tokens with expiration and signatures.
• Optimistic locking on inventory reservations.
• Idempotency keys on checkout and payment operations.
• OpenTelemetry traces, Prometheus metrics, structured logs, and Sentry.

Post-purchase: AI-powered service

The same thread handles tracking, returns, refunds, exchanges, and warranty claims. Rules and NSR guardrails ensure eligibility and caps; workflows trigger labels, refunds, restock, and accounting entries automatically.

Business outcomes

What is next

The roadmap expands predictive analytics, voice commerce, image-based search, deeper personalization, and B2B features (quote-to-order, contract pricing, EDI). Outcome-based pricing keeps incentives aligned as automation grows.