What happened

Zalando has launched an AI shopping agent. CNA reported it on 3 November 2026, and the report frames the launch against pressure on the retailer’s margins. I could not get past the headline facts, so I will keep to those. The point I want to make does not depend on the details.

The claim, and the test I apply to it

The claim behind a launch like this is that an AI agent will help a retailer that is being squeezed. So I ask the usual questions. Is there a need? Yes. Margin pressure in online fashion is real, and anything that moves it deserves attention. What actually changed? Shoppers get a chat that helps them find and choose products. What did not change? Every cost line that squeezes the margin in the first place.

At the chat level, every AI chatbot is almost the same. A shopper asks what goes with a jacket, and the agent answers well. Every store’s chat can do that now. It is pleasant, and it may lift conversion a little, but recommending is the cheapest thing an agent can do. It touches no cost that the finance team is worried about.

Where the margin leaks in fashion

Anyone who has run an online fashion store knows the shape of the problem. A shopper is unsure of the size, so they order two and send one back. Or they order one, it does not fit, and it comes back. Each return means shipping both ways, someone opening the parcel, checking it, putting it back into stock, and sometimes marking it down because the season has moved on. Add the support load around it: where is my order, how do I return this, when do I get my money back.

Those are the lines an agent has to touch if it is going to pay for itself. Size and fit sit at the top of the list, because getting them right stops the return before it exists.

What acting on that looks like

My position on automation is simple. It earns its keep when it checks, changes, books, charges or escalates inside the conversation. Applied to fashion, that means:

  • Check. When a shopper asks about a size, the agent looks at what they bought before, which sizes they kept and which they returned, and tells them which size in this brand matches what worked for them.
  • Change. If the shopper picked the wrong size and the order has not shipped, the agent changes it in the order system right there, instead of letting it ship and come back.
  • Book. If a return has to happen, the agent arranges the pickup in the same chat and tells the shopper when the refund lands.
  • Escalate. When something falls outside what the business allows, the shopper and the order details go to a person together, so nobody starts from zero.

None of this is a smarter answer. It is the agent reaching into the order system and the delivery system and doing something a person used to do. That is where the money is.

The same question for a smaller shop

Zalando can afford to launch, measure and adjust. A small fashion label in Singapore selling on Shopify and answering customers on WhatsApp cannot afford an experiment that only makes the chat nicer. So the test for the owner is the same one I apply to Zalando. Which of my costs can this agent change while the customer is still talking to me?

If a shopper messages at 11pm saying the dress arrived and the size is wrong, a chat that apologises and explains the returns policy has changed nothing. A chat that checks the order, confirms the right size is in stock, swaps it, and books the collection of the wrong one has removed a chunk of the return cost and kept the sale. This is the layer I built AeroChat on: it connects to the store’s data systems and backend processes so it can perform actions, not just answer questions. The recommendation part is the easy bit. The connection to orders, payments and delivery is the work.

What a business should actually do

Write down the three or four costs that squeeze your margin. For most fashion stores that is returns, refunds, delivery and support time. Then, before you buy or build any agent, ask one question of each cost line: can the agent change this inside the conversation, and what system does it need to be connected to for that to happen?

If the honest answer is that it recommends products and answers questions, treat it as a chat and budget it like one. If it can check the order, change the size before dispatch, and book the return when it must happen, then it is touching the margin, and you can measure that. Compare the return rate on orders where the agent handled sizing against orders where it did not, and judge it on that number.

Source: CNA Business: Retailer Zalando launches AI shopping agent as margins squeezed