Every time you linger on a product page, add something to your wishlist, or abandon a basket, you leave a trace. On a platform the size of Zalando, those traces accumulate across tens of millions of shoppers and feed directly into decisions about what gets bought from brands, in what volumes, and sometimes what gets designed in the first place. The feedback loop between your behaviour and the clothes that appear on site is real, consequential, and almost entirely invisible to the people inside it.
Key takeaways
- Zalando's AI-powered platform turns individual browsing signals into probabilistic demand forecasts at the level of a single article in a single market.
- Demand forecasting in fashion is genuinely hard: seasonality, fast trend cycles, and a vast product range all work against simple prediction.
- The same data infrastructure that personalises your homepage also informs buying volumes — meaning consumer attention shapes supply, not just the reverse.
- Regulation is beginning to catch up: Zalando is designated under the EU Digital Services Act as a very large online platform, which imposes new transparency requirements on its recommender systems.
- Merchandising analytics tools help brands and retailers read similar signals on their own terms, giving smaller players access to the same kind of feedback logic.
What does Zalando actually know about you?
When you browse Zalando, the platform records far more than your purchases. Clicks, dwell time on a product image, zoom gestures, wishlist additions, size selections that never became orders, and returns all generate data. Across an active customer base that reached an all-time high by the end of 2024, these signals aggregate into something closer to a real-time map of collective desire than a simple sales ledger.
This is not unusual for large e-commerce platforms, but the scale matters. A single shopper's wishlist tells you little. Fifty million shoppers' wishlists, filtered by region, age cohort, price sensitivity, and the week of the season, start to look like a demand signal that no traditional buyer's instinct can replicate.
How does browsing data become a buying decision?
The path from your click to a purchase order is neither instant nor direct, but it is traceable.
Step 1: Signal collection. Every interaction is logged — not just purchases but the full browsing session. Which images did you look at longest? Did you check the size guide? Did you return the item and, if so, why?
Step 2: Demand forecasting. Zalando's engineering team models demand prediction as a time-series forecasting problem, working at the level of an individual article in each market for any given day. The platform uses a deep recurrent neural network to produce probabilistic forecasts — meaning rather than predicting a single number, it estimates a range of likely outcomes and their probabilities. The team has described the challenges openly: fashion articles turn over fast, trends move unpredictably, and the sheer diversity of the catalogue makes each forecasting problem slightly different from the last.
Step 3: Buying and inventory decisions. Those probabilistic forecasts feed into decisions about how many units to order from a brand or supplier, which sizes to weight more heavily, and when to reorder or mark down. A product that generates high wishlist volume but low conversion might signal a price sensitivity problem. One that converts fast in a specific region might prompt a localised restock.
Step 4: The loop closes. What gets stocked shapes what you see. Personalised homepages and search rankings surface items the algorithm believes you are likely to buy — which in turn generates more signal data, which refines the next forecast.
Why is fashion forecasting so difficult?
Demand forecasting is a solved problem in many retail categories. Fashion is different for several reasons that are worth naming clearly.
- Short article life. A dress style may be on site for one season. There is little historical sales data to learn from before the window closes.
- Trend sensitivity. A colour or silhouette can go from niche to everywhere in weeks, driven by a single cultural moment. Models trained on last season's data may not capture this.
- Size complexity. A single style might come in twelve size variants, each with its own demand curve. Misjudging the size ratio means stockouts in popular sizes and excess in others — a problem that drives both lost sales and unnecessary production.
- Return rates. Fashion has among the highest return rates in e-commerce. A product's apparent sales figure is not the same as its real demand until returns are accounted for.
These are not hypothetical problems. They are the reason that even sophisticated platforms carry unsold stock, and why overproduction remains one of fashion's most persistent sustainability issues.
What does this mean for what gets made?
This is the part of the loop that most consumers do not see. Zalando operates both as a retailer (buying stock outright from brands) and as a marketplace (where brands sell directly to consumers through the platform). In both cases, the data it holds shapes upstream decisions.
For brands selling through the platform, Zalando's aggregate demand signals — which styles are trending, which sizes are selling through fastest, which price points are converting — are increasingly shared as part of the commercial relationship. A brand that sees its parka performing strongly in Scandinavia but slowly in Southern Europe can adjust its next production run accordingly. One whose size 38 sells out in two weeks while size 42 lingers can recalibrate its size ratio for the following season.
This is where the feedback loop touches production, not just inventory. When enough platforms share this kind of signal with their supplier network, the aggregate effect is a gentle but real pressure on what gets designed and manufactured. Consumer attention, mediated by platform data, becomes a soft instruction to the supply chain.
How are brands and retailers reading these signals themselves?
Not every brand has Zalando's data infrastructure. For merchandising teams at mid-size retailers and brands, purpose-built analytics tools fill a similar role. Style Arcade offers a buying and planning workspace that brings together sales, stock, and inventory data across channels, with visual range planning, size curve analytics, demand forecasting, and reorder recommendations. The goal is to give buying teams the same kind of feedback logic — what is selling, in what sizes, at what rate — without requiring a data science team to extract it.
The underlying principle is the same whether the platform is continental or boutique: reading what shoppers actually do, rather than what buyers predict they will do, produces more accurate orders and less waste.
What is still unsolved?
Data-driven merchandising is genuinely better than pure intuition at predicting stable demand. It is less good at several things that matter.
Novelty. A genuinely new product has no historical signal. The algorithm cannot tell you how a silhouette no one has seen before will perform. Human buyers still carry most of the risk on newness.
Cultural shifts. A trend that emerges from a subculture, a film release, or a political moment can invalidate months of training data almost overnight. Recurrent neural networks learn from the past; they do not anticipate ruptures.
Sustainability trade-offs. Optimising for sell-through reduces overstock, which is a real environmental gain. But the same logic can also accelerate the pace of newness — more styles, shorter runs, faster turnover — which has its own environmental cost. The data tells you what sells; it does not tell you what should be made.
Transparency to shoppers. Most consumers have no clear picture of how their browsing shapes what they see next, let alone what gets produced. The McKinsey State of Fashion report has consistently flagged consumer trust and data transparency as growing concerns for the industry. Shoppers are beginning to ask questions that platforms are not yet fully equipped to answer.
What does regulation say about this?
Zalando is designated as a very large online platform under the EU Digital Services Act, which means it faces obligations that smaller platforms do not. The DSA framework for very large online platforms requires systematic risk assessment of recommender systems, including their potential societal impact, and mandates that users be offered at least one recommendation option not based on profiling. This does not dismantle the data loop, but it does introduce a layer of accountability and a requirement to make the logic of recommendation systems more legible — to regulators, and eventually to users.
For fashion specifically, the implications are still being worked out. A recommender system that surfaces products based on your browsing history is also, implicitly, a system that shapes demand. Whether that constitutes a systemic risk under the DSA is a question regulators are actively examining.
The bigger picture
The feedback loop between shopper behaviour and fashion production is not new. Buyers have always tried to read the market. What is new is the granularity, the speed, and the scale at which that reading now happens. A platform with tens of millions of active customers, running probabilistic forecasts at the individual article level across dozens of markets, is doing something qualitatively different from a buyer reviewing last season's sell-through reports.
That is neither straightforwardly good nor bad. It reduces some waste. It can also accelerate churn. It personalises your experience in ways you may find useful or may find unsettling. And it connects your attention — your two seconds on a product image — to a production decision in a way that was not possible a decade ago.
Knowing the loop exists is the first step to thinking about what you want from it.
FAQ
Does Zalando share my browsing data with the brands whose products I look at? Zalando shares aggregate demand signals — which styles are trending, which sizes are moving — with brand partners as part of its commercial relationships. Individual-level browsing data is governed by its privacy policy and EU data protection law; personalised data is not simply handed to third-party brands.
Can my wishlist actually influence what gets restocked? Collectively, yes. Wishlist data is one of the signals that feeds demand forecasting. A product with high wishlist volume relative to purchases can flag price sensitivity or fit uncertainty, which may influence reorder or markdown decisions.
What is demand forecasting and why does it matter for sustainability? Demand forecasting is the process of predicting how many units of a product will sell, in which sizes and markets, over a given period. Better forecasts mean fewer units ordered speculatively, which reduces unsold stock — one of fashion's significant sources of waste.
What is the EU Digital Services Act and does it affect how Zalando uses my data? The DSA is EU legislation that imposes additional obligations on very large online platforms, including requirements around recommender systems. Zalando, as a designated platform under this framework, must assess the risks of its recommendation systems and offer users alternatives to profile-based recommendations.
Do smaller brands have access to the same kind of data feedback loop? Not at the same scale, but purpose-built merchandising analytics tools give smaller teams access to comparable logic — connecting sales, stock, and size data to buying decisions in a way that reduces guesswork and overordering.
Further reading
- Zalando Engineering: demand forecasting and mathematical optimisation in e-commerce
- EU Digital Services Act: very large online platforms
- McKinsey State of Fashion