Considerations: The retailer has failed the serial returner
Why banning serial returners is a human-centred design failure which will only intensify with AI powered discovery
A few weeks ago, I read a piece in the Financial Times about a woman who had been banned from Net-a-Porter for returning too many items. I’ve been ruminating on it since. Not because the story shocked me, but because the online retail experience has become so blasé, so disinterested in the consumers it claims to service.
The serial returner is not behaving untowardly. She is behaving exactly as the system designed her to. The retailer built the conditions: frictionless returns as a competitive differentiator, recommendation engines and editorial photography optimised for click and purchase rather than fit and longevity, and behavioural nudges to add more products to basket. The customer responded rationally. And then she got banned for it.
And she’s not alone: online apparel return rates run at 20–40%, with some luxury fashion segments seeing rates as high as 50%.1There is a term for this in human-centred design which I learned during my years of working in NHS service transformation: you do not blame the user for using the system in the way you built it to be used. You redesign the system.
What Net-a-Porter did instead is the equivalent of engineering a door that opens inward, watching people walk into it, and fining them for not knowing which way to lean.The returns crisis is not a customer behaviour problem. It is a discovery failure. The customer is returning because she bought the wrong thing. The wrong thing got bought because the discovery process failed to give her enough information to make a considered decision.
The escalation is this: AI-powered discovery is coming to luxury fashion, and it is going to make this problem significantly worse before the industry addresses it clearly. When the recommendation engine stops being a human editor, however algorithmically assisted, and becomes a machine optimised for conversion, the gap between what gets pushed and what actually serves the consumer will widen. Banning the returner now, before that gap grows, is penalising someone for structural and commercial problems the industry created. And quite frankly, I think it’s lazy of retailers.
Where is the investment in understanding the individual consumer, the drivers behind their previous purchases which resulted in returns, the reasons why the consumer returned the items? What does the data show the retailer about this consumer, are there themes and trends of returned items? Why aren’t retailers expanding their use of AI into more quantitative and qualitative analysis of their customers at individual and aggregate levels? Industry research from the Retail Technology Show found that almost a quarter of UK retail professionals warned that shoppers who return the most are often the highest spenders and that banning them risks losing some of the retailer’s most loyal and valuable customers.2 Bain & Company’s 2025 research on agentic AI in retail found that even as generative AI reshapes how consumers find and evaluate products, around half of shoppers remain cautious about letting AI agents autonomously handle purchases from start to finish.3 Consumers themselves are uncertain about AI-led discovery. Retailers are moving faster than consumer trust allows.
I fear that AI discovery, sold to consumers as a personalised experience is actually diluted behaviour nudges learned by the LLM based on aggregate data. The result? Not servicing the consumer on an individual basis therefore an increase in returns.
The human is at the centre of fashion, style and taste. The human should be at the centre of discovery: understanding the consumer’s physicality, lifestyle and context, how an item sits on the body, how it feels and the emotional response it triggers. AI discovery cannot empirically provide that to consumers. So how are retailers planning to restructure their returns policies and customer bans in the age of AI discovery?
The answer is not a stricter returns policy. It is not more friction at the returns desk. It is a deeper understanding of the consumer before the purchase happens. And here is where I want to be precise because digital, data and technology are part of the solution. But only when they are built to solve the right problem. The danger with AI in retail, as with many technology implementations I watched in the NHS, is the difference between a solution designed around a genuine human need and a solution looking for a problem to justify its existence. McKinsey has noted that the scaling of generative AI in retail is unique because its use cases involve direct interactions with consumers, and that even a 1% margin of error could result in millions of customer-facing mistakes.4 The worst outcome is not that AI fails to reduce returns and fails to provide a great, personalised customer experience. If AI reduces returns by making it harder to return, it suppresses the symptom while the underlying design failure compounds invisibly underneath.
So what does good actually look like for the retailer? Two things, neither of them radical.
The first is richer product information. The standard luxury e-commerce product page is designed to make the garment look beautiful. It’s not designed to help the customer understand whether the garment will work for them specifically. How does it sit on a body that is not a sample size? How does the fabric behave after a day of wearing? What is the weight of the leather, the opacity of the silk, the structure of the shoulder? These are the questions a customer asks in a good boutique and cannot ask online. Some retailers have gestured toward this with model height and size information. Some use short videos to show how the items move on the model. Some use inclusive models to showcase the items on different body sizes. But it’s not enough. The gap between editorial fantasy and physical reality is precisely where returns are created. Technology could close this gap: detailed fabric data, fit notes from real bodies across size ranges, videos of the garment in motion, inclusive sized models, but only if retailers are willing to invest in it as a customer service tool rather than an aesthetic one.
The second is treating returns data as insight rather than cost. If a customer is returning at volume, that is not a problem. It is information. It tells you they are engaged; they’re buying, they’re trying, they care about getting it right. The customer who returns seven dresses and keeps the eighth is not a liability. She is telling you, in the clearest possible language, exactly what she is looking for. An AI trained not on conversion data but on returns data: on what customers kept, what they sent back, and the gap between the two, would be a genuinely useful tool. It would ask: given what this customer has kept and what she has returned, what is she actually looking for? And that profile can be built longitudinally based on insights, and actually offer a truly personalised customer experience based on needs and wants. That is a fundamentally different application of the technology than the recommendation engine that got us here. It requires retailers to reframe the returns problem entirely: not as a cost to be minimised, but as a dataset that, read correctly, could transform the discovery and long-term customer experiences.
The most considered thing Net-a-Porter could have done when that woman hit her returns threshold was look at what she kept and ask why. Instead they sent a ban. That is not a technology failure. That is a failure of curiosity about the customer, the most expensive failure a luxury retailer can make.
References
1. NRF 2025 Retail Returns Landscape, National Retail Federation, October 2025.
2. A third of shoppers say retailers shouldn’t ban serial returners, Retail Technology Show research
3.Agentic AI in Retail: How Autonomous Shopping Is Redefining the Customer Journey, Bain & Company, November 2025.
4.LLM to ROI: How to Scale Gen AI in Retail, McKinsey & Company, August 2024
