Skip to main content
Support is the most common use case for Lookfor. By building a support-focused MAS, you can automate the majority of repetitive customer inquiries (order status, returns, refunds, shipping questions, and policy lookups) and only escalate complex cases to humans.

​
What you’ll build

A multi-agent system with specialized agents for each support domain, connected to one or more channels.

​
Step 1: Build your Knowledge Base

Before creating agents, prepare the data they’ll reference:
  1. FAQs: Write 10 to 20 FAQs covering your return policy, shipping times, sizing guide, payment methods, and store policies.
  2. PDFs: Upload your full return/refund policy, shipping guide, and any product manuals.
  3. Products and Collections: These sync automatically from Shopify.

​
Step 2: Create a Support MAS

  1. Go to Multi-Agent Systems > Create.
  2. Choose the MAS type matching your channel (Storefront, Email, Instagram, or Facebook).
  3. Add these agents:

​
Triage Agent (entry agent)

Role: Route every conversation to the right specialist. Instructions: 
You are the triage agent for {shop.name}. Your only job is to
identify the customer's intent and route to the right specialist.
Do not answer questions directly.

- Order-related (status, cancellation, tracking, returns, refunds):
  hand off to Orders Agent.
- Product questions (specs, availability, recommendations):
  hand off to Product Agent.
- Shipping questions (delivery times, methods, costs):
  hand off to Orders Agent.
- Everything else (policies, store info, general):
  hand off to General Agent.
Model: GPT-4.1 Nano (fast routing, no complex reasoning needed). Tool Choice: Required (forces a handoff on every turn).

​
Orders Agent

Role: Handle all order-related inquiries. Instructions: 
You handle order-related questions for {shop.name}.

When a customer asks about an order:
1. Ask for their order number or email.
2. Use Get Customer Orders to find their orders.
3. Use Get Order Details for the specific order.
4. Provide status, tracking, and next steps.

For cancellations: Confirm with the customer before using Cancel Order.
For address changes: Use Update Order Shipping Address.
For returns: Use Create Return and explain the process.
For refunds: Hand off to Refund Agent.
Tools: Get Customer Orders, Get Order Details, Cancel Order, Update Order Shipping Address, Create Return. Handoffs: Refund Agent, Live Agent.

​
Refund Agent

Role: Process refunds after the Orders Agent has gathered details. Tools: Refund Order, Get Order Details. Model: GPT-4.1 (refund decisions need careful judgment).

​
General Agent

Role: Answer FAQ and policy questions. Tools: Get Related Knowledge Source, Web Search. Model: GPT-4.1 Mini.

​
Live Agent

Role: Escalate to a human when the AI can’t resolve. Tools: Connect to Live Chat.

​
Step 3: Connect to your channels

  1. Go to Settings > Channels and assign the MAS to your preferred channels.
  2. Toggle Auto-reply on.
  3. Set a response delay of 2 to 5 seconds for chat, or 30 to 120 seconds for email, for a natural feel.

​
Step 4: Set up workflows

Create workflows to complement your MAS:
  • Assign to AI: The essential workflow. Use the Assign to AI Assistant template or create one with the Assign to AI action.
  • Priority escalation: If the message contains “urgent” or “complaint”, set priority to Critical and add an “Escalated” label.
  • AI labeling: Use the AI Label action to automatically categorize conversations (Orders, Shipping, Returns, General).
  • VIP routing: Route high-value customers to your best support agent.

​
Step 5: Monitor and iterate

  • Check the Dashboard for your AI automation rate. Aim for 70%+ within the first week.
  • Review Traces for conversations the AI handled poorly.
  • Create Reports for specific issues.
  • Check MAS Stats to identify slow tools or agents.
  • Refine agent instructions, add FAQs, and adjust tools based on what you observe.
NATPAT achieved a 91% resolution rate using this pattern. Read the case study.