The returns conversation in ecommerce has always been about friction reduction. Faster labels. Simpler portals. Fewer clicks between “I want to send this back” and “your refund is on its way.” The entire returns technology ecosystem, from Loop to Narvar to Parcel Lab, has been built around making that consumer-facing experience smoother.
That work matters. But it is solving for a world where a human places an order, receives a product, decides it is not right, and initiates a return through a portal designed for humans. Every assumption in the current returns infrastructure starts with a person making a decision at a screen.
Agentic commerce breaks those assumptions. When an AI agent places an order on behalf of a consumer, sometimes without the consumer actively choosing the specific product, the return scenario changes in ways that most operations teams have not yet thought through. The questions are not hypothetical. They are arriving alongside the first wave of AI-assisted purchasing, and the brands and 3PLs that start building answers now will have a structural advantage over those that wait for the problems to show up in their support queue.
The Questions Yet to be Asked of Agentic Commerce Returns
Start with the most basic one: who initiates the return?
In a traditional transaction, the consumer decides to return a product and takes action. They visit a returns portal, select a reason, print a label, and drop the package. The merchant processes the return, updates inventory, and issues a refund. Every step is triggered by a human decision.
When an AI agent places the order, the consumer may not have selected the specific product. They may have expressed a preference (“I need a moisturizer for dry skin under $40”) and the agent chose the specific SKU. If the product does not meet the consumer’s expectations, is the return initiated by the consumer, by the agent, or by some automated dissatisfaction signal? The answer determines which system triggers the return flow and what data accompanies it.
Next: who is the merchant of record?
This question is straightforward when a consumer buys directly from a brand’s website. It gets complicated fast in agentic scenarios where an AI agent aggregates products from multiple merchants into a single purchase experience. If the agent places orders across three different brands and the consumer wants to return one item, does the return go back to the brand, to the agent’s platform, or to a third-party fulfillment provider the agent selected? The merchant of record determines who processes the refund, who absorbs the shipping cost, and who receives the returned inventory.
Then: where does the returned product go?
In standard returns, the destination is typically the original fulfillment location or a returns processing center. The consumer sends it back, the warehouse receives it, and the item either goes back into sellable inventory or gets dispositioned. But if the AI agent selected a fulfillment path that the brand did not control, the return destination may not be obvious. If the agent routed the order through a marketplace fulfillment service or a third-party logistics provider, the return path may not mirror the outbound path. Someone has to own that routing decision, and today, nobody does.
How Returns Flow Back Into Inventory
The inventory implications of returns are already one of the messiest problems in ecommerce operations. In an agentic world, they get worse.
When a returned item arrives at a warehouse, it needs to be inspected, graded, and either restocked or dispositioned. If it goes back into sellable inventory, the available quantity across every connected channel needs to update. This is the same real-time inventory challenge that makes outbound operations difficult, but in reverse, and with additional variables.
In a standard return, the operations team knows which SKU is coming back, from which order, to which location. The return authorization carries the data needed to match the inbound item to the original transaction. In an agentic return, some of that data may be missing or structured differently. The AI agent may have its own order identifier that does not map cleanly to the merchant’s order number. The return reason may be expressed in natural language rather than selected from a dropdown. The condition of the item may need to be inferred rather than declared.
Each of these gaps creates a processing delay at the warehouse. Processing delays create inventory inaccuracies. Inventory inaccuracies create overselling or underselling on every connected channel. The cascade is the same one that makes outbound inventory hard, but the inputs are less predictable.
The Customer Experience Problem
Returns are already a leading driver of customer service volume. In the prestige beauty segment, where products are often bought as gifts or based on shade matching, return rates can run significantly higher than the ecommerce average. For apparel, returns on certain categories exceed 30%.
In a traditional return, the customer knows what they bought, why they are returning it, and what they expect to happen. The communication cadence is clear: confirmation of return request, label generation, tracking, receipt at warehouse, refund processed. Each step has a corresponding notification.
In an agentic return, the communication layer gets complicated. The consumer may not know exactly what they ordered if the AI agent made the selection. They may not know which merchant to contact. They may expect the AI agent to handle the return process, but the agent may not have the infrastructure to do so. The result is a customer who is confused about who to talk to, uncertain about where to send the product, and frustrated by a process that was supposed to be frictionless.
This is where the customer experience degrades fastest. The promise of agentic commerce is convenience. Returns that are harder to navigate than the original purchase break that promise.
What the Returns Tech Stack Needs to Handle
The current returns technology ecosystem was built for a specific workflow: consumer initiates return through a branded portal, return is processed by the merchant, inventory and financials update accordingly. Platforms like Loop, Narvar, and Parcel Lab excel at making this workflow fast and consumer-friendly.
What they were not built for is a scenario where the return may be initiated by an agent rather than a consumer, where the merchant of record may be ambiguous, where the return destination may not match the fulfillment origin, and where the data accompanying the return may be structured in a format the merchant’s systems do not expect.
The integration layer matters here. A returns platform that connects to a brand’s Shopify store and ERP handles the standard flow well. But when returns need to be routed based on fulfillment path rather than merchant identity, when inventory updates need to account for items in transit from multiple return origins, and when financial reconciliation needs to match refunds to orders that were placed through an intermediary, the returns platform alone cannot solve it.
This is where the order operations layer becomes critical. The same platform that manages outbound order routing, inventory computation, and financial data flow needs to handle the reverse flow as well. Returns are not a separate system. They are the outbound process running backward, with additional complexity at every step.
A Framework for Thinking About Returns in Agentic Commerce
Tim Morse, VP of Revenue at Pipe17, framed the problem in a recent internal strategy session: “How do you even handle returns in an agentic world? Who is the merchant of record? Where do the returns go back to? How are they flowed back into inventory? And what is the customer experience?”
Those four questions form a useful framework for any brand or 3PL thinking about how to prepare:
Merchant of record clarity. Before an AI agent can place an order on your behalf, the rules governing returns need to be explicit. Who processes the refund? Who pays for return shipping? Who receives the physical product? These rules need to be codified in the agent’s operating parameters, not left to be resolved after the return is initiated.
Return routing logic. Just as outbound orders need intelligent routing based on inventory, proximity, and cost, returns need routing logic that accounts for where the product was fulfilled from, where the merchant wants it returned to, and what disposition rules apply. This logic should live in the order operations layer, not in the returns portal.
Inventory reconciliation at the speed of commerce. Returned items that sit in a processing queue for days before updating available inventory are lost revenue. The same real-time inventory computation that prevents overselling on the outbound side needs to account for inbound returns, adjusting channel-level availability as soon as a return is received and graded.
Customer communication ownership. Someone needs to own the communication with the consumer throughout the return process. In an agentic scenario, this may mean the AI agent communicates return status, or the brand communicates directly, or both. The handoff between agent and merchant needs to be defined before the first return is initiated.
Where This Goes
Returns in agentic commerce are not a problem that needs to be solved today. They are a problem that needs to be architected for today so that the solution is in place when the volume arrives.
The brands and 3PLs that treat returns as part of their order operations infrastructure, not as a separate workflow managed by a separate system, will be better positioned to handle whatever agentic commerce throws at them. The ones that wait until AI-initiated returns are causing support ticket spikes will be building the plane while flying it.
The returns problem has always been an operations problem disguised as a customer experience problem. Agentic commerce just makes the disguise thinner.
