Returns involve the physical movement of goods; refunds involve money. AI handles the logistics part (return labels, drop-off instructions, tracking the parcel back) and the money part separately. Mistakes happen when the workflow treats them as one action, confirming a refund before the goods have actually been returned, or holding a refund longer than the policy required because the return tracking failed silently.
A customer wants to return a product. The AI issues a return label and tells the customer they will receive their refund once the return is processed. The customer reads "once the return is processed" as a question of when, not whether. The AI's workflow then has to know when the warehouse confirms receipt, trigger the refund, and tell the customer. The handoff between logistics and finance is where the error usually lives.
What people in the field are saying
Gorgias and similar Shopify-native tools have built specifically around the return-then-refund pattern in e-commerce. The pattern is so common that the integration with the order management system is the difference between an AI that does this well and one that does it badly.
What does AI do well in returns?
Generate the return label inside the conversation. Provide drop-off instructions tailored to the customer's location. Track the parcel back and update the customer at each step (received, processed, refunded). Handle the routine variations (wrong item, damaged in transit, customer changed mind).
What does it do badly?
Confirms a refund timeline that depends on a logistics step the AI did not actually verify. Loses the case in the gap between the logistics system saying "received" and the finance system saying "refunded." Fails to chase the customer when the return label was issued and the parcel never moved.
Where do customer complaints come from?
The AI promised a refund within X days. The X days passed. The customer chased. The AI said "your return was received, the refund is being processed," which is true and unhelpful. The actual problem (the finance system needs three more business days for cross-border transfers) was never communicated. The customer's experience is that the AI confirmed a promise and then evaded it.
What is the practical pattern?
Treat returns and refunds as two linked workflows, not one. The AI manages both, with a clear audit trail of each step. Communicate the actual timeline including the finance step ("you'll see the refund within 5-7 business days after receipt; cross-border refunds can take an extra 3 days"). Chase if any step takes longer than expected.
What is the smallest first step?
Test the return-to-refund flow end to end for the most common return reason. Time each step. Communicate the timing honestly in the AI's response. Customers handle long timelines better than they handle vague ones.
Related: use case 4: processing a refund, use case 2: order status, and the glossary explainer on first contact resolution.