RTFP

Zalando has established a dedicated Trust & Safety (T&S) team to manage content moderation. The team’s primary responsibility is to design comprehensive policies and guidelines to ensure Zalando’s standards of platform trust and safety are met. Complementing this initial moderation, all content once published may be reported by customers via the Notice & Action mechanism if still deemed inappropriate or harmful. These two distinct layers of moderation processes ensure that only appropriate content is published. By maintaining these stringent content moderation practices, Zalando not only adheres to regulatory requirements but also upholds the integrity and safety of its platform. Zalando content moderation function is overseen by the T&S team who also have policy ownership over the enforcement guidelines and Community Guidelines. T&S, Legal and Compliance teams work closely on the creation of these policies and guidelines and ensuing effective implementation. We have a hybrid approach to support T&S with an internal T&S team and an external vendor team that carries out the content moderation work to support operations. T&S is part of Zalando’s product organisation and works closely with Legal, Privacy, Compliance and Public Policy on policy updates and escalations. Significant policy changes, review of high‑risk cases, and fundamental changes to our safety processes follow a documented review process through our T&S Steering Committee. Our outsourced vendor also supports the training and quality of our teams ensuring we have the correct QA approach and framework for our policy enforcement.

Zalando uses a third party platform to moderate content. The vendor platform utilizes AI models for content moderation. These models automatically approve non-violating content and reject violating content. Content flagged as potentially violating is escalated to human reviewers. Zalando works with another vendor that provides content moderation services by humans. All content is moderated before it is made public on the Zalando platform.

This is done based on projected volumes of content, turn around times and market coverage. In launching new products and features we assess the user adoption rates and project potential volumes by quarter. We have a service level agreement with our moderation partner and in order to meet those SLAs we assess how many people we would need to meet those SLAs with the projected volume of content. We also look at the markets we will launch in to ensure we have appropriate time and language coverage.

Zalando has in-house resources with various expert know-how (Expert Teams), as well as contracted a vendor for human moderation. The team members of the vendor are hired and selected via our vendor operations team based on language proficiency, regional/cultural familiarity, and professional experience in Trust and Safety operations, compliance, or investigations. The roles include frontline content moderators, senior reviewers and QA. Where vendors are used, our contracts require baseline vetting, background screening consistent with local law, and language-level assessments and resilient testing in line with the wellness standards for T&S work. As for the Expert teams, their qualification encompasses: 9 years of Zalando-specific content moderation experience; Bachelors of Science in Biology, Industrial Microbiology, and Clothing & Textile Technology; Bachelor in Fashion Design; Bachelor's Degree in Chemical Engineering; Master of Arts in International Management; Industrial Engineering degrees; Master in Retail and Consumer Management, Data Management, and Translation; Fully qualified lawyer (German law); Bachelor in Pharmacy and Master of Business Administration (MBA); LL.M., MSc in Political Science, and Master of Science in International Business Administration (with Data Science & Decision Support focus); Master of Science in International Business, Sustainable Management, and Technology; Paralegal certification; Master of Arts in Art History and Comparative Literature; and specialized Trust & Safety professionals with extensive platform-specific content moderation and operations experience. Our moderation ecosystem spans EQF Levels 4 through 7. Frontline vendor operations and QA structures align with Levels 4 and 5, ensuring robust technical proficiency. These are bolstered by our in-house Expert Teams, whose academic and professional credentials map to Levels 6 and 7, providing the high-level specialized knowledge required for complex compliance and legal investigations.

We monitor automated performance using standard quality indicators such as precision (how often flags are correct), recall (how often violations are detected), and observed false positive and false negative rates based on Quality Assessment sampling. We also use operational indicators such as moderator disagreement rates, appeal overturn rates, and periodic spot checks to identify drift and systematic errors, and to adjust prompts, automated action thresholds, and workflows over time.

We identify and deactivate harmful content from our platform using a combination of people and technology to flag content, review content, and enforce our community guidelines. We use a third-party tool to deploy AI/ML models to flag potential community guideline violations to support content moderation (prioritizing human review) and to automate enforcement of violations. Our scanning approach uses a combination of LLM models, with prompt-driven detection for text, image, and contextual signals, and pre-trained moderation models for image and video flagging. These automated tools return structured labels and confidence scores that we use to either immediately reject the content or route items into moderation queues, to prioritize review, and to recommend an enforcement outcome. 77% of AI-supported enforcement actions are confirmed by trained human reviewers, with 727 pieces of content automatically removed.

We apply safeguards to reduce risk and ensure accountability. This includes human in the loop review for the majority of enforcement decisions, conservative confidence thresholds for any automated takedowns, and Quality Assurance processes such as sampling, duplicate review, and performance monitoring dashboards. Access controls and audit logs restrict who can view or act on cases, and data handling follows privacy and security by design practices such as encryption in transit and at rest, role based access control, and defined retention controls. Users can be provided with reasons for enforcement and appeal routes where applicable, and policy and threshold changes are documented and tested to limit unintended automation impact.

Automated means are applied for clearly defined purposes: 1) Initial triage and prioritization so that likely violating or high risk content is reviewed faster. 2) Policy specific detection to support human decision making, including surfacing the likely policy reason and relevant model signals. 3) Automation of low ambiguity cases only when confidence is high and the enforcement logic is deterministic, otherwise the item is escalated to human review. 4) Quality and reporting support, including measurement of volumes, trends, and enforcement consistency for transparency reporting and compliance workflows.

In October 2025, Zalando launched new optional public customer profiles, which offer users in all Zalando markets a personalised space to save content (ie. existing content on Zalando), follow brands and creators, and share with other users on Zalando. Only the names of the Boards, the user profile handles and their descriptions are created by users. Content moderation at the provider's own initiative refers to moderation by Zalando of profiles and boards. Zalando’s own initiative moderation capabilities are a mix of AI and human moderating systems along with guidelines and policies. Upon submission for publication, all content immediately undergoes an automated moderation process guided by our established community guidelines. Only a subset of the moderated content is then reviewed by human moderators. The outcome of this initial moderation is either a success, leading to the content's immediate publication, or a rejection. In cases of rejection, the user is promptly notified regarding the specific violation, and the content is consequently not made publicly available.

The vendor provides well‑being support including access to employee assistance programs (EAP), on‑call wellness resources including trained physiologist onsite and content desensitisation tooling where feasible (e.g., blurring, audio muting). As a general principle, the vendor will ensure moderators go through rotations away from high‑severity queues when exposed to more egregious content, and workload monitoring. Managers are trained to identify signs of distress, and staff can opt out of certain content categories subject to business needs. Zalando is provided with regular reporting from the vendors to ensure that they are providing regular check‑ins, offer confidential counselling, and implement break protocols to mitigate exposure to potentially harmful content. As for the inhouse Expert teams, resources include: General Zalando Wellbeing services and employee assistance initiatives; Product, Content and Brand (PCB) Guidelines; dedicated Legal team support for complex decision-making; P&C Chat for real-time collaboration; Inhouse inspection units and chemical/physical product testing at accredited external laboratories; preemptive capacity support and cross-team reassignment models for high-inbound volume management.

The external moderators complete a foundation layer onboarding trust and safety training programme including policies, prohibited content categories, user safety, privacy, data protection, and confidentiality. They will also undertake any specific role‑specific training on review tools and escalation and market specific content if required. Ongoing training includes policy refreshers, edge‑case calibration, QA feedback loops, bias awareness, and regional/cultural context is part of the training program that is a hybrid between Zalando and the outsourced vendor. Zalando runs scheduled calibration sessions to align decision thresholds across teams and track policy comprehension via assessments. Training content is updated as required with any new legislation or when policy changes occur. As for the Expert teams, they go through the following training: Digital Services Act (DSA) Process Training including Notice & Action and Salesforce Hypercare; Intellectual Property (IP) protection and WIPO training; Brand Protection operational panels; Training on Ratings and Reviews policy and Community Guidelines; Traceability and compliance management systems for environmental claims; Quality Management Training; comprehensive Onboarding and Standard Operating Procedures (SOP) for all article types.