Transforming Compliance with Machine Learning

EPAM Partners with Zalando to Build a Scalable, Automated EPR Compliance Solution Leveraging the Power of Databricks and AWS

At a glance

CLIENT

Zalando

STRATEGIC PARTNERS

Amazon Web Services (AWS)
Databricks

SERVICES

Artificial Intelligence
Data & Analytics

INDUSTRY

Retail

Zalando, one of Europe’s leading online fashion platforms, was looking to elevate its Extended Producer Responsibility (EPR) compliance process. Faced with the challenge of manually managing EPR regulations across an eight-digit number of product articles, Zalando needed a scalable, automated solution to tag and categorize its products. Leveraging Databricks and AWS, EPAM partnered with Zalando to develop a sophisticated machine learning (ML) product that combines automated tagging capabilities with intelligent oversight mechanisms, ensuring both accuracy and adaptability — now and in the future.

SUSTAINABLE FASHION

Addressing Complex Regulatory Challenges

With over 52 million active customers across Europe and more than 6,000 brands in its portfolio, Zalando holds a unique position in the European fashion ecosystem. Zalando is ready to step up and embrace its role as an enabler — acting as a catalyst, convener and connector to drive progress toward a more responsible fashion industry. Zalando, like other manufacturers, is accountable for the entire lifecycle of its products, particularly when it comes to waste management and recycling, through the EPR policy.

In 2024, Zalando’s manual EPR compliance process was becoming increasingly difficult to manage. With millions of products requiring precise classification for waste management regulations, the company recognized the need to enhance efficiency and reduce the risk of errors, ensuring smooth operations and compliance across multiple markets.

Zalando approached its longstanding partner EPAM for its deep data and ML expertise to develop a solution that addressed these challenges. Together, we developed an integrated ML product that transformed how Zalando approaches EPR compliance.

Building a Scalable EPR Solution

EPAM’s team developed an expandable, extensible and configurable system leveraging AWS cloud infrastructure that automatically assigns EPR tags. The solution is built on top of ML-based data pipelines in Databricks, where each model is precisely trained to tag items according to EPR legislation based on product attributes, display names and descriptions — delivering classification accuracy that surpasses human capabilities while maintaining the flexibility to adapt to evolving regulations.

An essential part of the solution is the EPR Control Plane, which provides dynamic reconfiguration capabilities to enable real-time adjustments to categories, priorities and compliance rules. This helps Zalando rapidly respond to regulatory changes across different countries while seamlessly integrating new product categories into its compliance framework.

The new EPR solution delivered significant improvements in four key areas:

Making the EPR process more transparent

Improving data quality, accuracy and reliability

Reducing manual validation efforts

Scaling the EPR classification process

Key Solution Features

Automatically assigns EPR tags to all product articles and any new products that become live on the site

Enables the reviewing or reassigning of ML-generated tags when a product has a very low probability of EPR prediction

Allows users to create or edit a new EPR category for a particular country based on a change/update in legislation

Enables a version control feature that records if and why an EPR tag has been edited or updated
Allows users to associate multiple countries with a single EPR category to efficiently reuse models for the same legislation across countries and improve maintainability

Enables users to assign priorities for EPR categories and override EPR tags for specific cases

Allows filtering and sorting of the data in tables for a better user experience

Provides the ability for the model to be retrained with new data to improve accuracy and/or accommodate potential changes in legislation