Let an AI Agent Handle Returns: a Policy It Can Follow

An AI agent can handle returns when you give it a written policy with exact windows, conditions, exclusions and hand-off rules: it checks the order against the policy and opens the return, and a teammate approves the refund. The agent does the back-and-forth, and a person decides when money goes back.

Returns are one of the best jobs to give an AI agent, and one of the easiest to get wrong. The questions are frequent, the answers depend on a few clear facts (when it was delivered, what it is, what condition it's in), and customers want an answer now, not tomorrow. This guide shows how to write a policy an agent can actually follow, with a template you can copy, and where a person should step in.

• The facts are in your store: Order date, delivery date, items and price are all in the order. An agent with order lookup doesn't need to ask the customer for them.
• The rules are written down: Most return decisions follow the policy on your website. Edge cases are the exception, and those go to your team.
• Speed matters: A customer who gets a clear answer and a return started in two minutes is far less likely to dispute the charge with their bank.

A person on your team fills gaps in a vague policy with judgment and memory of last time. An agent fills them with its best reading of the words. So write the policy as if for a new hire on their first day: specific numbers, named exceptions, and what to do when something doesn't fit.

• Use numbers, not adjectives: "Within 30 days of delivery," not "within a reasonable time." "Unopened and unused," not "in good condition."
• Say which date counts: Order date, shipping date or delivery date. Delivery date is fairest to the customer and easiest to check from tracking.
• Name the exclusions: Final sale items, gift cards, personalized products, opened consumables, intimate apparel. List them by product type or tag the agent can see.
• Say what to offer, in order: For example: exchange first, then store credit, then a refund to the original payment method.
• Say who pays return shipping: And whether that changes for damaged or wrong items.
• Say when to bring in a person: Damaged items, missing packages, large orders, repeat returners, angry customers and anything the policy doesn't cover.
• Give the reason: An agent handles new cases better when it knows why a rule exists. "We don't accept opened supplements because we can't resell them" helps it answer a question you never wrote down.

Copy this into your knowledge or policy rules and fill in the brackets. Keep it in one place so every agent, on every channel, answers the same way.

• Window: Returns are accepted within [30] days of delivery, using the delivery date in tracking. Orders with no delivery scan: hand off to the team.
• Condition: Items must be [unused, unwashed, with tags]. Consumables must be [unopened].
• Not returnable: [Final sale items, gift cards, personalized items, opened consumables].
• Options, in this order: [Exchange for another size or color] → [store credit for the full amount] → refund to the original payment method.
• Return shipping: [Customer pays return shipping], except for damaged, defective or wrong items, where [we pay].
• Damaged or wrong items: Ask the customer for a photo of the item and the packing slip, then hand off to the team with the photo and order number.
• Subscriptions: A subscription order follows the same rules. If the customer wants to stop future orders, offer to [skip or pause] first, then cancel if they still want to.
• Always hand off: Orders over [$300], a customer's [third] return in [90] days, missing-package claims, chargeback threats and any request outside this policy.
• Refunds: The agent never promises a refund amount or date. It tells the customer the return is open and the team will review the refund.

Most returns are honest. A few patterns are worth a human look, and an agent should hand them off instead of deciding:

• Frequent returners: Several returns in a short period, especially of worn or used items.
• Missing items or empty boxes: Claims that a package arrived empty or a part was missing need the shipping weight and a person's judgment.
• "Delivered but not received": These are carrier questions first, and often turn into a reship rather than a return.
• Pressure: Threats of a chargeback, a review or legal action. A calm person handles these better, and your instructions can say so.
• Requests to refund somewhere else: A refund to a different card, account or person than the one that paid.

• The return is opened: When the customer asks, the agent opens the return and tells the customer what happens next, without promising an amount or date.
• Your team reviews it: A teammate checks the return and issues the refund in your store, the same way they do today.
• Unclear cases come with notes: When the agent hands off, the conversation shows up in your team's queue with the reason and a summary. See hand off to a teammate.

If you use Shopify's self-serve returns, customers can also request a return from their order status page, and you approve or decline each request in your Shopify admin based on your policies (Shopify Help Center, checked October 19, 2026). An agent and self-serve returns work well together: the agent answers the questions around the return, and both paths end with a person approving the money.

• Which products are returned most, and for what reasons? Fix the product page, size chart or packaging before the next order.
• Which returns did the agent hand off, and why? If the same question keeps coming up, add the answer to the policy.
• How many returns became exchanges or store credit instead of refunds?
• Are any customers returning far more than others?

In Concierge, an operator agent can run this every week from the refund analysis playbook and post it in Slack or the Finn app.

Concierge handles returns on chat, phone, text, WhatsApp and email. It checks the customer's identity with a sign-in code, looks up the order, applies your policy, opens the return the customer asks for, and passes anything unusual to your team. Refunds stay with your team. Every step is saved in the conversation, so you can see what the agent did and why. Concierge Starter is $20 a month with 20,000 credits. Pro is $199 a month, or $1,990 a year, with 199,000 credits. More credits cost $1 per 1,000, phone and text charges are billed as used, and there are no per-seat fees. See pricing.

Glossary: human in the loop, approval workflow, standard operating procedure and knowledge base. Read next: how to keep a human in the loop, AI escalation and human handoff and store credit vs points. Choosing a tool? See the best AI customer service agents for Shopify.

Some tools let it. Concierge doesn't: the agent opens the return, and a teammate approves the refund, so money only moves when a person says so.

The return window and which date counts, item conditions, items that can't be returned, the options to offer in order, who pays return shipping, and a clear list of cases to hand to a person.

An AI agent should verify the customer first. Concierge asks the customer to confirm a sign-in code before it shares private order details or changes their account.

An AI agent shouldn't promise a refund, and your policy can say so. Tell it to confirm the return is open and that the team will review the refund.