Returns Are Predictable (And That's Good)

Every return request follows the same pattern:

1. Customer says they want to return something
2. You check if it's within the return window
3. You verify the reason (defective? wrong item? changed mind?)
4. You issue an RMA number or return label
5. You process the refund or exchange when the item arrives

This flow is the same whether you sell shoes, software, or kitchen appliances. The details change. The process doesn't.

Predictable processes are the easiest to automate.

What to Automate

Return intake and classification. When a customer says "I want to return this," classify the intent (return_request, exchange_request, warranty_claim) and ask the two questions that matter: order number and reason for return.

Auto-response: "We can help with your return. To get started, can you share your order number and the reason for the return? Here are the common reasons: [defective/wrong item/changed mind/other]."

Return eligibility check. If you can connect to your order system via API or webhook, check whether the order is within the return window automatically. "Your order #12345 was placed 12 days ago. Our return window is 30 days, so you're eligible. Here's how to proceed: [link to return portal]."

If the order is outside the return window: "Unfortunately, your order #12345 was placed 45 days ago and our return window is 30 days. For exceptions, I've forwarded your request to our returns team."

Warranty claim routing. Warranty claims need more documentation (photos of the defect, proof of purchase, product details). Auto-respond with a checklist: "To process your warranty claim, we need: 1) your order number, 2) a photo of the defect, 3) a description of the issue. Reply with these and we'll review within 1 business day."

Route the claim to your warranty team with all the details attached. They review and decide — the automation handles the intake.

RMA number generation. If your system supports it, generate the RMA number automatically and include the return shipping address: "Your RMA number is RMA-2026-4521. Ship the item to [address]. Once we receive it, your refund will be processed within 5-7 business days."

What NOT to Automate

Exception decisions. Customer is 2 days past the return window? A human should decide whether to make an exception. Automation follows rules. Exceptions need judgment.

Refund processing. The actual refund (crediting the customer's payment method) should be triggered by a human confirming receipt of the returned item. Don't auto-process refunds before verifying the return.

Warranty denial. If a warranty claim is denied (damage caused by customer, warranty expired), a human should communicate that. An automated "warranty denied" message with no explanation will generate angry follow-ups.

The Flow for E-Commerce

Here's the full automated return flow:

Customer: "I want to return the shoes I bought last week." AI classification: return_request (confidence: 94%) Auto-response: "Happy to help with your return. Can you share your order number?"

Customer: "Order #38291" System: Checks order database. Order placed 8 days ago. 30-day return window. Eligible. Auto-response: "Your order #38291 is eligible for return. Start the process here: [return portal link]. You'll receive a prepaid shipping label by email."

If the customer doesn't have the order number: Auto-response: "No problem. You can find your order number in your confirmation email, or I can look it up. What email address did you use to place the order?"

If the order isn't eligible: Auto-response: "Your order #38291 was placed on [date], which is outside our 30-day return window. I've forwarded your request to our returns team for review. They'll respond within 1 business day."

The Numbers

For an e-commerce store processing 100 returns/month:

Manual processing:

• 100 returns × 10 minutes each = ~17 hours/month
• Staff cost: 17 hours × $20/hour = $340/month

Automated processing:

• 70 returns auto-handled through the flow: 70 × $0.25 = $17.50
• 30 returns needing human judgment: 30 × 5 minutes = 2.5 hours
• Staff cost: 2.5 hours × $20/hour = $50
• Total: $67.50/month

Savings: $272.50/month. And the customer gets their answer in seconds instead of hours.

The biggest win isn't the cost savings — it's the customer experience. A customer who gets their return label in 10 seconds shops with you again. A customer who waits 2 days for a return response shops elsewhere.