"How do I automate returns with AI" is really six questions: which decisions in a return can a machine take today, which still need a person, in what order to hand them over, how to write the rules so the machine takes them the way you would, how to tell a returns tool with an AI feature from an operation run by agents, and how to know it is working. This page answers them in that order. Pango's own capabilities are stated from its fact sheet; competitors are described only by their own descriptions; the one customer example is the published Switch Nails case study.
Short answer: Six decisions in a return can be automated today, in this order: eligibility (window, condition, final sale, country), the exchange offer (the right size, another product or store credit before a refund), the label or label-free drop-off and the choice of return carrier, the refund timing (on request, first scan, warehouse receipt or inspection), fraud and abuse flags, and the disposition when the item lands. A person is still needed for a disputed condition, a claim with photos that contradict each other, a high-value exception outside the rules, and anything the rules do not cover. Automate eligibility and the exchange offer first, because they decide the economics; refunds and fraud next; the warehouse last, because it needs the receiving connection. Write the rules in plain language, one policy at a time, and measure the share of returns that complete without a person, the exchange rate and the time to refund.
Free for DTC brands: Pango runs the shipping and returns audit on your own order and carrier data as part of a demo: fifteen numbers across checkout, carriers, warehouse handover, tracking and returns, the three that matter most for your operation, and the same orders shown running on one system. Book the free audit.
What can be automated today
| Decision in the return | What the automation does | What it needs to see |
|---|---|---|
| Eligibility | Applies the window, condition rules, final-sale items and the country's legal minimum per item, and tells the customer which items qualify | The order, the items, the policy per country |
| The exchange offer | Offers the right size, a different product or store credit before money back, and prices the difference | Live stock and prices |
| Reason capture and routing | Takes the reason and a photo, and routes a faulty item to a different outcome than a change of mind | The reason codes you report on |
| Label, drop-off and return carrier | Books the carrier, picks prepaid label or label-free QR code, chooses the cheapest or nearest option for that customer and country, and sets the return address | Carrier accounts, return addresses per warehouse or 3PL |
| Refund timing | Releases the refund, credit or exchange order at the point your rule names: on request, on the carrier's first scan, on warehouse receipt or after inspection, with a different rule for high-value or first-time customers | The return parcel's tracking and the warehouse receipt |
| Fraud and abuse flags | Counts returns per customer, flags serial returners and inconsistent photo evidence, and holds the case instead of refunding | Order history and the return record |
| Disposition | Records restock, refurbish or write off from the warehouse grade and updates stock | The warehouse scan |
What an autonomous return actually does, step by step, and why it needs the operation underneath it, is set out in autonomous returns: how AI agents run returns end to end. The difference between a rules-only portal and AI that decides per case is in AI returns management.
What still needs a person
Automation handles the routine cases, and the routine cases are most of them. A person is still needed for a disputed condition where the customer's photo and the warehouse grade disagree, a claim that mixes damage, delay and a change of mind, a high-value exception the rules did not anticipate, a customer the fraud flag has held who turns out to be a good customer, and any return in a country whose policy you have not written yet. The design goal is not zero people; it is that the person sees only the case, with the order, the tracking, the policy and the warehouse record already on one screen, and that their decision becomes a rule if it will recur.
The order to automate in
- Eligibility and the exchange offer. These two decide the economics: the first stops ineligible returns at the door, the second turns refunds into kept revenue. To increase the exchange rate, make the exchange the easy default, not a hidden option, offer the right alternative before you offer money back, present it the moment the customer starts the return, and remove friction like waiting for the item to arrive back; the method is in how to increase your exchange rate.
- The label and the return carrier. Once the portal books the carrier and sends the label or QR code, the return parcel is tracked like an outbound one and the refund clock has a scan to start from.
- Refund timing and fraud flags. With the return tracked, the refund rule can be tied to the first scan or the receipt, and the abuse limits can hold the exceptions.
- The warehouse. Receiving, grading and disposition close the loop, and they need the warehouse or 3PL connected, which is why they come last.
The customer sees the first two steps as the portal; what the portal has to do and how it connects to the warehouse and refunds is in returns portal software.
What the rules look like in plain language
The rules that drive the six decisions are short when written the way a returns manager says them. Examples of the shape, not of any one brand's policy:
- "Returns are accepted within 30 days of delivery, 14 days for sale items; final-sale and hygiene items are not returnable."
- "Offer the same item in another size first, then any product of equal or higher value, then store credit with a bonus, then a refund."
- "A faulty item with a photo is refunded on request; a change of mind is refunded when the warehouse receives it."
- "Refund first-time customers on the carrier's first scan; refund customers with more than three returns this year on inspection."
- "Use the cheapest carrier with a drop-off point within five kilometres; in Norway use the return address in Oslo."
- "Hold any return where the photo does not match the reason, and send it to a person."
A system that compiles sentences like these into workflows, rather than a settings page with checkboxes, is what lets the returns manager change the policy without a developer. The test to run in any demo is to change the window, the exchange order and the refund timing yourself.
A returns tool with AI, or an operation run by agents
Several vendors describe AI in returns by their own descriptions: AfterShip describes rules-based returns automation with instant refund on label scan; Loop Returns describes AI-assisted exchange recommendations and fraud signals; ReturnGO describes AI returns automation that auto-approves and recommends exchanges; Redo describes an AI concierge with action-taking workflows; Narvar describes agentic resolution across its post-purchase suite; Swap describes an AI agent that handles returns, exchanges and policy questions. The comparison that matters is not the label. It is whether the AI sees the whole return, the order, the outbound shipment, the return parcel's tracking, the policy and the warehouse receipt, or only the request form, because a decision taken on the form alone has to be checked by a person. The vendors by portal, rules, exchanges and pricing are compared in best returns management software, and the category in returns management software.
How Pango does it
Pango runs returns as one module of a post-purchase operations platform on the same record as checkout delivery, shipping, tracking and the warehouse. Natural-language return and exchange rules are compiled into per-merchant workflows, exchanges can be to any product in the store rather than only a variant swap, and labels, cross-border documents and refund logic are set per merchant and per country. Because the return shares a record with the outbound shipment and the warehouse, the agent that takes the decision can see the tracking, the receipt and the policy at once, and a case outside the rules is escalated with all of it attached. Switch Nails, a Nordic press-on nail brand, runs its returns this way: 99% of returns run fully self-serve, 19% of returns now stay with the brand as an exchange or store credit, up from zero before Pango, and 33% of exchangers place another order against 20% of refund-takers; the full account is in the Switch Nails case study.
What to measure
- Share of returns completed without a person. The first number to ask a vendor for, and the one that shows whether the automation is real.
- Exchange rate. The share of returns that end as an exchange or store credit rather than a refund.
- Time to refund. From request to money, by refund rule.
- Cost per return. Labels, carrier, handling and the software, per return.
- Return rate by product and reason. The number that feeds back into the catalogue and the size guide.
- Exceptions per hundred returns. How many cases reach a person, and why; each recurring reason is a rule to write.
The bottom line
Automating returns with AI is a sequence, not a switch: eligibility, exchange offer, label and carrier, refund timing, fraud flags, warehouse. Each step needs the system to see one more part of the operation, and the rules only work if the returns manager can write and change them in plain language. Measure the share of returns that complete without a person and the exchange rate, and treat every recurring exception as the next rule.
