How Zalando uses AI: the assistant, personalisation and size advice

KEY TAKEAWAYS Summary by the editors

1. Zalando uses AI in three customer-facing areas: a conversational shopping assistant built on OpenAI models, a personalised discovery feed and size advice based on return data, brand measurements and computer vision.
2. Zalando states that about one third of its overall returns are size-related and that its size advice has reduced size-related returns by 10%.
3. According to an OpenAI customer story, switching the Zalando Assistant to GPT-4o mini was followed by 23% more product clicks and more than 40% more wishlist additions.
4. Zalando has not disclosed the assistant's effect on conversion, basket value, returns or margin, so its commercial impact remains unproven in public data.
5. The transferable lesson is to start where proprietary data already exists, such as return reasons and fit feedback, and to treat the language model as an interface layer over one's own catalogue.

Zalando uses AI in three customer-facing places: a conversational shopping assistant built on large language models, personalised discovery such as its AI-driven feed, and size advice that combines return data, brand measurements and computer vision. The company has published a few results, notably a 10% reduction in size-related returns through size advice, but most commercial impact figures remain undisclosed.

What has Zalando built?

Zalando, the Berlin-based online platform founded in 2008, reported 62 million active customers and more than 7,000 brand partners across 29 markets in June 2024. On top of that platform it has layered several AI features:

• Zalando Assistant: a conversational tool, first launched in 2023 in four German- and English-speaking markets, that lets customers describe what they need in natural language and receive product suggestions.
• Trend Spotter: a weekly updated view of emerging trends in six European cities (Berlin, Paris, Milan, Antwerp, Stockholm and Copenhagen), later integrated into the assistant.
• Discovery feed: a personalised feed of products, brands and content, launched in six countries in July 2025 and extended to 16 more markets in October 2025.
• Size advice: size recommendations based on purchase and return history and, since July 2023, a feature that predicts body measurements from two smartphone photos.

Zalando's 2026 AGM management board report shows the assistant moving closer to the transaction: customers can add an item to the bag and pick a size, ask what is already in their bag, ask which products go with it and go straight to checkout from within the conversation.

Why did Zalando invest in AI?

Two problems run through Zalando's public statements. The first is discovery. With a very large assortment, a search bar and category filters make it hard for customers to express a need such as an outfit for a particular occasion. Zalando reported that queries typed into the assistant were three times the length of searches in the search bar, which suggests customers use it for more complex requests. The second is returns. Zalando states that about one third of its overall returns are size-related, which makes fit a direct cost and sustainability issue.

A strategic motive sits above both. When Zalando extended the discovery feed in October 2025, co-founder Robert Gentz described it as a pivotal moment in transforming Zalando from a primarily transactional platform into an inspirational one. AI-driven personalisation is the mechanism behind that shift, because a feed only works if it is relevant to each customer.

Read also

How adidas uses AI: from design archives to generative content

How does it work (data, models, process)?

The assistant runs on OpenAI models. According to a customer story published by OpenAI, Zalando moved the assistant from GPT-3.5 to GPT-4o mini over two weeks, initially routing half of its traffic to the new model, and used component-specific evaluations and improved few-shot prompting to raise answer quality. The language model interprets the request and writes the reply; the products it shows come from Zalando's own catalogue and customer data.

Trend Spotter is built on customer desire signals rather than sales: searches, likes and additions to the cart, according to Zalando. The size engine draws on a wider mix of data: item measurements supplied by brands, customers' return history and stated return reasons, fit feedback on earlier orders, the sizes customers enter in their profile, sizes of items bought elsewhere, input from fitting models who try on garments, and computer vision that flags sizing patterns across similar items from the same brand.

What results has Zalando reported?

Zalando and its model provider have published a small set of figures. None is audited, and most measure engagement rather than profit:

• Size advice: Zalando says its size advice offerings have reduced size-related returns by 10%.
• Assistant usage: more than 500,000 customers had used the assistant by June 2024, with about four back-and-forth exchanges per conversation on average.
• Assistant engagement: OpenAI's customer story reports a 23% increase in product clicks, more than 40% more wishlist additions and 5% fewer recommendations rated unhelpful, with traffic scaling twelvefold without a significant rise in cost.
• Feed reach: more than 25 million unique users had interacted with the feed since launch, according to the 2026 AGM report.

Zalando has not disclosed the assistant's effect on conversion, basket value, returns or margin, nor what the language models cost to run. Some engagement figures come from a vendor publication, which readers should weigh accordingly.

What are the limits and open questions?

The following is editorial analysis rather than company disclosure. Engagement is not the same as incremental sales: more clicks and wishlist additions can also mean customers browse longer before buying the same items. Without controlled results on conversion and returns, the assistant's business case cannot be judged from outside.

The size figure needs careful reading. If the 10% reduction applied to all size-related returns, it would equal roughly three percent of total returns, because size accounts for about a third; Zalando has not stated that calculation. Size advice also depends on brands supplying accurate measurements, which varies across thousands of partners.

Building the assistant on an external model creates dependency on a provider's pricing, latency and model changes, although Zalando's migration in two weeks shows the risk can be managed with good evaluation. Finally, personalisation and body measurement use sensitive personal data, so consent, transparency and data minimisation are ongoing obligations rather than one-off design choices.

Read also

How Nike uses AI: direct-to-consumer data and generative product creation

What can other fashion companies learn?

1. Start with a costly, measurable problem where proprietary data already exists. Return reasons and fit feedback were Zalando's raw material for size advice.
2. Invest in product data first. Size advice is only as good as the garment measurements brands supply.
3. Treat the language model as an interface over your own catalogue, so the model can be swapped without rebuilding the product.
4. Build component-level evaluations before scaling; they are what made a fast model migration possible.
5. Measure business outcomes, not only engagement, and decide in advance which metric will justify further investment.

Frequently asked questions

Does Zalando use ChatGPT?

Zalando's assistant runs on OpenAI models. According to OpenAI, it was first built on GPT-3.5 and later moved to GPT-4o mini. Product suggestions draw on Zalando's own catalogue and customer data rather than on the general model alone.

What is the Zalando Assistant?

It is a conversational shopping tool inside the Zalando app and website that lets customers describe what they are looking for in natural language. It suggests products, includes trend information from Trend Spotter and, according to Zalando's 2026 AGM report, can add items to the bag and lead to checkout.

How does Zalando's size advice work?

It combines brand measurements, customers' purchase and return history with return reasons, fit feedback, profile sizes, fitting-model input and computer vision. Since July 2023 customers can also receive recommendations based on body measurements predicted from two smartphone photos.

Has AI reduced returns at Zalando?

Zalando says its size advice has reduced size-related returns by 10%, and that about one third of all its returns are size-related. It has not published the effect of the conversational assistant on returns.

GuideThe complete guide to AI in fashion e-commerce, marketing and retailRead the complete guide