Use Case Intelligence RSS Search use cases Search
OpenAI Retail Europe & UK

Zalando

Marketing Personalization

Zalando needed to scale its shopping assistant from a few markets to 25 multilingual markets while improving personalization, instruction following, and recommendation quality without sharply increasing cost.

Read the original story Explore all use cases
12x traffic growth without spending significantly more
23% increase in product clicks
40%+ increase in products added to the wishlist
5% decrease in unhelpful recommendations

Reported impact

Business outcome

Zalando increased product clicks by 23%, boosted wishlist additions by more than 40%, scaled the assistant to all 25 markets, and expanded traffic 12x without spending significantly more after moving to GPT-4o mini.

  • Cost Reduction
  • Customer Experience
  • Revenue Growth
  • Scalability

Solution

Zalando rebuilt its shopping assistant around GPT-4o mini and a more granular evaluation framework so personalization could scale across Europe. The first assistant version used GPT-3.5 in a handful of German and English markets, but struggled with nuanced fashion requests, multilingual instruction following, and broad rollout economics. Working with OpenAI, Zalando introduced component-specific evals that score routing and response generation separately and improved few-shot prompts with stronger examples of good and bad recommendations. Once those diagnostics were in place, the team migrated traffic to GPT-4o mini in stages, first moving half the traffic in two weeks and then finishing the rollout across 25 markets. The new stack produces localized recommendations in multiple languages, handles detailed event- and season-specific requests better, and gives Zalando a scalable way to keep improving the assistant as conversational commerce use cases deepen.

Data flow

A shopper request enters the Zalando Assistant, which routes it through the evaluated assistant stack and sends generation tasks to GPT-4o mini. Component-specific evals and few-shot examples monitor routing and recommendation quality so localized responses can be improved before and after rollout across each market.

Solution architecture

4 components · 4 layers
  1. Compute
    • GPT-4o mini Powers multilingual, instruction-following recommendation and conversational shopping tasks at lower cost and latency.
  2. Serving
    • Zalando Assistant Provides personalized product discovery and recommendation experiences for shoppers across Zalando markets.
  3. Orchestration
    • few-shot prompts Teach the model what strong and weak recommendations look like so quality stays aligned with user expectations.
  4. Governance
    • component-specific evaluations Measure routing and response generation separately so the team can improve weak parts of the assistant stack.

Architecture clues

  • Assistant supports local languages and cultural contexts across 25 markets
  • Component-specific evaluations isolate routing and response generation quality
  • Few-shot prompts provide explicit examples of strong and weak outputs
  • Migration moved 50% of traffic to GPT-4o mini in two weeks

Evidence from the source

"By transitioning from GPT-3.5 to GPT-4o mini, Zalando was able to scale its traffic 12x"
"The Assistant now performs seamlessly across all 25 of Zalando's markets"
"The new framework included component-specific evaluations to test individual parts of the system"
"Zalando worked with OpenAI to develop the Zalando Assistant"
Transparency Evidence & methodology 100/100 reference value · 87% confidence · 1 source

Review how this reference was scored and trace each normalized claim back to its documented source.

01

Scoring methodology

How scores are calculated

Customer Reference Value measures usefulness as a customer reference. Evidence Confidence measures the quality and traceability of available evidence, not independent verification.

Customer reference value100/100
  • Problem clarity20/20
    Measures whether the customer's operational problem is specific enough to compare.
  • Implementation depth25/25
    Rewards named components, data flow, integrations, and architecture detail.
  • Outcome specificity20/20
    Measures whether the reported benefit is concrete rather than a generic claim.
  • Quantified impact25/25
    4 metrics describe measurable impact.
  • Deployment scale10/10
    Rewards evidence of a live or scaled deployment and meaningful integration scope.
Evidence confidence87%
  • Source authority20/25
    The main evidence is an official vendor-published customer story.
  • Claim traceability21/25
    10 of 12 key fields link to a named source.
  • Direct evidence20/20
    4 direct source quotes retained.
  • Date provenance8/10
    The publication date comes from publisher-controlled metadata.
  • Source transparency8/10
    1 source reference document publisher, URL, authority, retrieval, and supported fields.
  • Implementation specificity10/10
    Rewards source-backed components, data flow, quantified impact, and a specific outcome.
02

Traceability

Sources & provenance

Field coverage shows which facts each source supports. Dates are shown only when documented by the source.

  1. OpenAI

    Zalando customer story

    First-party
    https://openai.com/index/zalando/
    Source type
    Vendor-published customer story
    Published
    Dec 11, 2024
    Supports
    • Customer
    • Business problem
    • Solution
    • Products
    • Solution components
    • Data flow
    • Integrations
    • technical area
    • use case category
    • Business outcome
    • outcome category
    • Metrics
    • Architecture
    • evidence quotes
    • Publication date

Date evidence

Not separately sourced Industry, Region, Company size. Treat these normalized attributes as lower-confidence until a dedicated source is added.

Related

Similar use cases

See all like this
OpenAI Nov 25, 2025 Target Retail United States Target wanted to help shoppers who start with broad, conversational questions and do not know exactly what they want, reducing the need to sift through large search result sets before finding affordable, relevant… multiple times a day Customer Service AutomationMarketing Personalization ChatGPT · Target account · Target app in ChatGPT OpenAI May 5, 2025 Lowe's Retail North America Lowe's needed to give DIY shoppers and store associates project-specific guidance across thousands of products and 1,700 stores while also improving merchandising, forecasting, and supply-chain decisions without making… 1,700 stores Enterprise Knowledge SearchMarketing Personalization Mylow · Mylow Companion · OpenAI's APIs OpenAI Nov 13, 2024 Mercari Retail Japan Mercari needed to reduce the effort and guesswork sellers face when creating listings, because poor descriptions, missing attributes, and post-sale friction limited conversion and slowed marketplace growth. Over a few hundred AI-assisted listings are created per minute Marketing PersonalizationSales Automation AI Listing Support · GPT-3.5 Turbo · GPT-4