Workflow Examples

Overview

Shopigent Returns works in two modes:

  1. Chat Mode — Talk to an AI agent (Claude, Codex) and it handles returns via MCP
  2. Automation Mode — The agent runs on a schedule and processes returns automatically

Workflow 1: Daily Return Processing

Goal: Every morning, process all pending returns automatically.

Step 1: Find pending returns

The agent calls:

list_returns(status: "PENDING")

Result: Returns a list of all returns waiting for review.

Step 2: Analyze each return

For each pending return, the agent calls:

analyze_return(returnId: "xxx")

Result: A recommendation (approve/review/deny) with confidence score and policy match.

Step 3: Auto-approve eligible returns

For returns that match an auto-approve policy:

approve_return(returnId: "xxx", issueLabel: true)

Result: The return is approved, refund is executed in Shopify, label is generated, customer gets an email.

Step 4: Flag for review

For returns that don't match any policy: → The agent creates a report for the merchant to review manually.

Chat Example

Merchant: "Process all pending returns for today."

Agent: "Found 5 pending returns. Analyzing..."

  • #BZ-2026010 — Maria P. — $45.00 — ✅ Auto-approved (matches Standard policy)
  • #BZ-2026015 — John D. — $299.00 — ⚠️ Needs review (over $200 policy limit)
  • #BZ-2026016 — Anna K. — $89.00 — ✅ Auto-approved
  • #BZ-2026017 — Peter M. — $150.00 — ✅ Auto-approved
  • #BZ-2026018 — Sofia R. — $500.00 — ❌ Denied (outside 30-day window)

"3 approved with refunds, 1 denied, 1 needs your review."

Workflow 2: Fraud Investigation

Goal: Check all new returns for fraud signals before processing.

The agent calls:

  1. list_returns(status: "PENDING") → get all pending returns
  2. For each: check_fraud(returnId) → run fraud detection

Detection signals:

Signal What it checks Risk
High-value return Items over $1,000 Medium
Frequent returner 3+ returns in 30 days High
Amount mismatch Return amount doesn't match order High

Chat Example

Merchant: "Check for fraud on today's returns."

Agent: "Running fraud checks on 3 pending returns..."

  • Order #BZ-2026020 — Medium risk (high-value: $1,200)
  • Order #BZ-2026021 — 🚨 High risk (frequent returner: 4 returns this month)
  • Order #BZ-2026022 — Low risk ✅

"Recommend flagging #BZ-2026021 for manual review."

Workflow 3: Customer Support Automation

Goal: A customer emails about a return. The merchant asks the agent for an instant answer.

Chat Example

Merchant: "Customer Maria Papadopoulou wants to return a leather jacket from order #BZ-2026010. What's the policy?"

Agent: "Let me check..."

  1. list_returns(limit: 10) → finds Maria's return
  2. analyze_return(returnId: "xxx") → evaluates against policies

"Found it! Order #BZ-2026010 — Leather Jacket ($299.99) — 23 days ago.

  • Policy: Standard 30-day return
  • Eligible: ✅ (23 days ≤ 30 days)
  • Amount: ❌ ($299.99 > $200 auto-approve limit)
  • Recommendation: Manual review required"

"Since the jacket is over $200, I can't auto-approve it. Would you like me to approve it manually or let the customer know?"

Workflow 4: End-of-Month Analytics

Goal: Get a complete returns performance report.

Chat Example

Merchant: "Show me my returns analytics for this month."

Agent: "Let me gather the data..."

  1. list_returns() → get all returns
  2. list_policies() → get active policies

📊 Monthly Returns Report:

  • Total returns: 45
  • Auto-approved: 32 (71%)
  • Denied: 8 (18%)
  • Manual review: 5 (11%)
  • Total refunded: $3,240
  • Fraud prevented: ~$890
  • Avg resolution time: 2.3 hours

Top reasons: Sizing (40%), Changed mind (25%), Defective (15%) Most returned: Classic Leather Jacket (8 returns)

"Your auto-resolution rate is 71%. I'd recommend adding a policy for items over $200 to catch more auto-approvals."

Workflow 5: Bulk Approve with Labels

Goal: Approve 10 returns at once and generate labels for all of them.

Chat Example

Merchant: "Approve all pending returns under $100 and generate return labels."

Agent: "Finding pending returns under $100..."

Processing 6 returns:

  1. ✅ #BZ-2026025 — $34.99 — Approved + Label generated
  2. ✅ #BZ-2026026 — $89.00 — Approved + Label generated
  3. ✅ #BZ-2026027 — $45.00 — Approved + Label generated
  4. ✅ #BZ-2026028 — $22.50 — Approved + Label generated
  5. ✅ #BZ-2026029 — $79.99 — Approved + Label generated
  6. ✅ #BZ-2026030 — $12.99 — Approved + Label generated

"Done! 6 returns approved, $284.47 in refunds processed, labels sent to customers."

Automation Setup (Scheduled Tasks)

For fully automated processing, you can set up a daily cron job:

# Run every morning at 8 AM
0 8 * * * curl -X POST https://returns-app-production-8384.up.railway.app/api/mcp \
  -H "Authorization: Bearer YOUR_MCP_KEY" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"list_returns","arguments":{"status":"PENDING"}}}'

Or use an automation platform like n8n, Zapier, or Make to trigger MCP calls on a schedule.

Time Savings Calculator

Task Manual (per return) Automated
Review return request 5 minutes 2 seconds
Check policy eligibility 3 minutes Instant
Process refund 5 minutes 5 seconds
Generate label 3 minutes 3 seconds
Email customer 2 minutes Automatic
Total per return ~18 minutes ~10 seconds

For a store with 50 returns/month: Saves ~15 hours of manual work per month.