𝐖𝐡𝐞𝐧 𝐚𝐧 𝐀𝐈 𝐀𝐠𝐞𝐧𝐭 𝐁𝐮𝐲𝐬 𝐭𝐡𝐞 𝐖𝐫𝐨𝐧𝐠 𝐏𝐫𝐨𝐝𝐮𝐜𝐭, 𝐖𝐡𝐨 𝐒𝐡𝐨𝐮𝐥𝐝 𝐏𝐚𝐲 𝐟𝐨𝐫 𝐭𝐡𝐞 𝐑𝐞𝐭𝐮𝐫𝐧?

At Carrefour in Taiwan and China, we refunded without asking why.

Our belief was simple: most customers are honest. The many shouldn’t pay for the few. Trust you extend comes back.

My rule: no questions asked, as long as the product came back in good condition.

That worked because returns policies assumed something we rarely had to question:

The customer made the purchase.

AI shopping agents break that assumption.

Now one line in your returns report can hide three very different stories:

- The customer chose wrong.
- The agent chose wrong.
- The customer simply changed their mind.

Treat all three alike and you learn nothing.

Worse, tighten the policy for everyone and your honest customers pay for software errors. Keep absorbing every return and retailers may end up subsidizing the agents making those errors.

China offers an early glimpse of where this can go.

One decade ago, it put into law seven-day, no-reason returns for qualifying online purchases. Platforms later pushed further with “refund only,” allowing some buyers to keep the product and get their money back. Abuse followed, and major platforms pulled back.

What survived was attribution.

On Taobao, faster refunds can depend on the buyer’s history. The customer may face little friction, but the system still remembers who is transacting.

AI agents complicate that model because software is neither honest nor dishonest.

It is accurate or inaccurate.

And when it is inaccurate, retailers need to know whose error created the return.

This matters for more than cost.

As agents increasingly mediate discovery, selection and purchase, the return may become one of a retailer’s last direct moments with the customer.

Don’t waste it interrogating someone about a mistake their software may have made.

Resolve the customer’s problem first. Attribute the cause behind the scenes.

Your three moves:

1. Keep “no questions asked” for the person. Condition remains the test.
2. Add attribution to every return: customer choice, agent error, fulfillment error, product issue, or changed mind.
3. Require agent platforms to transmit the customer’s original instruction and the agent’s action. Once agent-caused returns are measurable, make cost allocation part of the commercial agreement.

Returns were designed as a trust mechanism.

In agentic commerce, they also need to become an attribution mechanism.

Trust the person. Verify the agent. Invoice the platform.

#AgenticCommerce #RetailStrategy #CustomerTrust #CustomerExperience #ChinaBusiness

Contact

bruno.gentil@sherpaconsultingasia.com

© 2026. All rights reserved.