You have probably seen the term agentic AI appearing more frequently in commerce coverage. Most of what gets written focuses on the buying side: AI assistants that help customers discover products, compare options, and complete purchases. That is the part that is easy to demonstrate.
The return side is less discussed, but for merchants it may matter more.
What "agentic" actually means
An agentic AI is one that can take actions, not just answer questions. Instead of telling a customer "you can return your order by visiting this link," an agentic AI can actually execute the return process on the customer's behalf: verifying the order, capturing the reason, checking eligibility, and resolving the case while the customer stays in the conversation.
The key difference from a chatbot is that the agent is calling real backend systems and producing real outcomes. Not a scripted flow. Not a handoff to a human. An actual resolution.
Why returns are a better use case than buying
Returns are structurally well-suited to agentic resolution for a few reasons.
First, they are high-friction for customers. Finding the return portal, re-entering order details, navigating reason dropdowns, waiting for approval: this process is genuinely annoying, and customers who find it too difficult often dispute the charge instead, which is worse for everyone.
Second, they are high-cost for merchants. A processed return involves reverse logistics, restocking or write-off, customer service time, and a refund. For preference-based returns, the customer changed their mind, the sizing was off, or the item was not what they expected, many of these costs could be avoided if the merchant had one earlier conversation with the customer before the return portal was opened.
Third, the logic is bounded. Unlike buying, where customer preferences are infinite, a return has a defined state: an order, a reason, a set of merchant rules, and a small number of possible outcomes. That structure is exactly what makes it automatable.
What an agentic return looks like in practice
In the live session we ran on houseofparfum.nl, the flow looked like this from the customer's side:
The customer told Gemini they wanted to return their order. Gemini asked for the order number. The customer provided it. Gemini listed the items in the order and asked which ones they wanted to return. The customer selected two. Gemini asked for the reason. The customer said "changed my mind." Gemini offered a small keep discount instead of a refund. The customer accepted. A real discount code was issued. A confirmation email arrived.
The entire exchange took under two minutes. No portal was opened. No support ticket was created. The merchant kept the revenue minus a small incentive cost.
What the merchant side looked like
From the merchant's side, KeepCard handled the verification and resolution logic. The agent called four endpoints in sequence: check eligibility, select items, submit reason, accept offer, and each one returned a deterministic response that the agent used to continue the conversation.
What merchants need to enable this
To run an agentic return flow with KeepCard, a merchant needs a connected Shopify or WooCommerce store, agentic returns enabled in the KeepCard dashboard, which generates a per-store API key, and an AI client or agent runtime that can call HTTP endpoints. The UCP Playground integration we used for the WooCommerce demo is one option. Any runtime that supports custom tool definitions can work.
The return logic itself, eligibility checking, reason classification, keep offer rules, fraud controls, is already configured in KeepCard. The agent does not need to know any of that. It just calls the API and follows the response.
The near-term trajectory
Shopify's native agentic storefronts for ChatGPT, Copilot, and Google AI Mode are already live for eligible stores. These channels will generate purchases. Some of those purchases will generate returns. The merchants who have a return layer that those same AI channels can call will handle those returns at a fraction of the cost of the merchants who do not.
That is the practical case for agentic returns. Not AI for its own sake, a cheaper, faster resolution for a problem that was already costing money before AI existed.