The Hidden Data Gap in Your Returns
Across the events and conversations of the last couple of months, one theme has come up consistently: the internet is being rebuilt around machines as first-class users. Our CTO Carlos Blanco has been distilling what that shift means in practice, and specifically what it means for fashion and home decor brands managing returns.
Here is his thinking.
The agent is already shopping
According to Cloudflare Radar, more than half of all traffic across Cloudflare’s network is now automated. Cloudflare sits in front of roughly 20% of the internet, so this is not a niche signal. Bots, crawlers, AI agents. The number shifted sharply in Q4 of last year and is continuing to grow.
Shopping assistants are no longer a concept being piloted in a lab. They are here, being used by real consumers to browse catalogues, compare options, and in some cases complete purchases on their behalf. A customer asking an AI to find them a sofa in a particular style, within a budget, and order it is not a far-fetched scenario. For many consumers it is already the default.
When an agent mediates the purchase, the relationship between the buyer and the item changes. The person who receives the product was not necessarily the one who chose it, evaluated it against alternatives, or decided it met their needs. The distance between the decision and the experience grows. And when that distance grows, return rates tend to follow.
Agentic browsing also introduces patterns that look nothing like human browsing. Agents access long-tail product pages that human traffic rarely reaches. They run multiple parallel sessions. They interact with product catalogues and inventory systems in ways those systems were not designed for. The traffic shape is fundamentally different, and most e-commerce infrastructure was not built with this in mind.
Returns face the same problem
Brands are asking the wrong question at the point of return.
Most return processes are built around a binary: did this item come back or not. Some go a step further and ask the customer to select a return reason from a dropdown. Sizing issue. Changed my mind. Item not as described. That data is better than nothing, but it is not very useful.
Customers do not always know why something did not work. And as agent-mediated purchases become more common, the stated return reason will matter even less. When a customer returns a jacket or a chair that an AI selected for them, the person filling in the return form may have had limited involvement in the original decision. There is a genuine legal question emerging around this: when an agent acts on your behalf and the outcome is not what you wanted, accountability for that outcome is genuinely unclear. Who decided? What information did they have? What were the parameters?
Brands processing those returns are left with very little signal to work with. A dropdown selection that does not reflect the actual decision process is not a useful input for improving anything.
The question that actually matters is: what is the condition of this item, and why did it come back? Those two things require evidence. They require a signal captured at the moment the item is received.
The simplest thing most brands are not doing
A photograph. At the point of return, before the item is touched or processed.
It sounds trivial. It is not.
A photograph at intake gives you an objective record of condition at the moment of return. It separates what the customer reported from what is actually true of the item. It makes condition assessment faster, more consistent, and less dependent on the individual judgment of whoever is processing the return that day. It creates an audit trail. And it gives you the raw material to do something much more valuable over time.
Most fashion and home decor brands skip this step entirely. The item arrives, a person looks at it, makes a call, and it moves on. No image. No structured record. No data.
The operational case is immediate
With image capture at return, condition assessment can be supported by computer vision. Rather than relying on a person to evaluate wear, defect type, and resale viability from memory and habit, the image becomes an input to a model that has seen thousands of similar items.
This matters for speed. It matters for consistency across warehouses and processing centres operating at different scales. And it matters for the decision itself: whether a garment is resold, repaired, donated, or recycled, or whether a piece of furniture goes back into inventory, gets refurbished, or is written off, is a consequential choice with real cost and sustainability implications. Getting it right consistently requires more than a quick look from a different person each day.
The structured data produced at this step, category, brand, material composition, condition grade, defect type and location, paired with return reason and time since purchase, is the foundation for better routing decisions. Not just today, but at scale.
The strategic case compounds over time
Every image captured is a labelled data point. And labelled, real-world data about returned fashion and home decor items is genuinely scarce.
AI systems are increasingly useful not when they have been trained on everything, but when they have been trained on the right things. The shift happening across AI applications right now is from bulk ingestion toward targeted, structured datasets that ground models in domain-specific reality. A model that understands what a lightly worn cashmere jumper looks like after six months versus eighteen months, or how a particular fabric defect presents across different weights, or how upholstery wear differs by material and usage pattern, is more useful than a general model with access to everything. That kind of specificity only comes from real operational data, captured consistently and structured in a way that makes it usable.
Brands that start capturing images at return now are building that dataset. Brands that do not are making a decision they will have to reverse later, at much higher cost, and with much less data to work with.
The gap between early movers and late movers here is not linear. Every return processed adds a data point. Every data point improves the next assessment. After two or three years of volume, that compounding effect becomes a genuine competitive asset that is very difficult to replicate from a standing start. The window to build this foundation is open now. It will not stay open indefinitely.
What CIRQUEL does
This is exactly the problem we built CIRQUEL to solve.
Our platform captures structured visual data at the point of return and uses computer vision to assess item condition consistently and at scale. Every item that moves through CIRQUEL generates a structured record: condition, defects, category, material, return context. That data feeds the routing decision immediately and contributes to a model that improves with every return processed.
We have validated this across both fashion and home decor, running pilots with brands in sportswear, footwear, accessories, furniture and home décor. The same principle holds across categories: the image is where the signal begins, and without it the rest of the process is guesswork at varying degrees of accuracy.
For brands, the result is faster and more accurate triage, better decisions about resale versus repair versus recycling, and a growing dataset that becomes more valuable as return volumes shift and as agentic purchasing introduces patterns that existing processes were not designed to handle.
The infrastructure for agents to shop already exists. The question is whether brands are building the data foundation to understand what comes back when those agents are buying at scale.
Starting with a photograph is not a small thing. It is where the signal begins.
Carlos Blanco is CTO and co-founder of CIRQUEL. If you are a brand, investor or operator thinking about returns, reverse logistics, or AI applications in physical supply chains, we would love to connect at cirquel.co.