Zalando

Objective

Help customers get inspired and find the items they like faster, at the consideration stage, to turn browsing into high-value interactions (like, add to basket) across all of Zalando's European markets.

The deployment

The Zalando Assistant is a conversational shopping assistant available to logged-in customers in the app and on the site, in their local language. The customer phrases a request in natural language, for example what they should wear for a birthday in November in Barcelona, and the assistant suggests outfits and items taking into account the place, the weather, and the occasion, as well as the section being browsed. Launched in beta for logged-in customers in October 2024, it was deployed from that date across Zalando's 25 markets, built on Zalando's in-house models and OpenAI's language models. On March 27, 2025, Zalando launched an enhanced version: a new visual identity, connection to the customer account for deeper personalization, and better use of the browsing context. On the technical side, the team moved the assistant from GPT-3.5 to GPT-4o mini, migrating 50 percent of the traffic in two weeks, which made it possible to extend the service to all markets and multiply traffic twelvefold with no significant extra cost. In the pilot phase, Zalando observed +40 percent high-value interactions such as a like or an add to basket; this figure is a pilot, not a generalized result. More than 2 million customers have used it since October 2024.

Results Proof B

Official Zalando release (T1 primary, corporate.zalando.com) quantifying adoption (more than 2 million customers), coverage (25 markets), and the pilot (+40 percent high-value interactions), consistent with the OpenAI customer story (T2, GPT-4o mini) which documents the stack and the scale-up (traffic 12x, recommendations deemed unhelpful -5 percent). Public and consistent figures but none consolidated into financial results, and the +40 percent is explicitly a pilot, hence a B level.

The internal detail is not public. Here is a proven approach that leads to the same result, to adapt to your stack.

1. 1
   Customer request customer

   The logged-in customer describes a need in natural language in the app or on the site (occasion, style, context).

2. 2
   Interpretation and retrieval AI

   The assistant interprets the request via the LLM, cross-referencing the customer account, the browsing context, and the catalog to select items and outfits.

3. 3
   Response and recommendation AI

   The assistant suggests outfits and items suited to the place, the weather, and the occasion, with the option to like or add to basket.

4. 4
   Oversight and model arbitration data team

   The teams track adoption, high-value interactions, and the share of recommendations deemed unhelpful, and arbitrate the underlying model (GPT-3.5 to GPT-4o mini migration) based on cost and quality.