Product descriptions:
adidas, the iconic German sports brand, has a legacy of innovation — from pioneering screw-in studs that revolutionized soccer cleats to pushing boundaries in performance, comfort and shopping experiences. To uphold this legacy and meet the needs of customers across 150+ countries, adidas needed a faster, smarter way to analyze product reviews and uncover customer sentiment. That’s why the brand partnered with Databricks to build a scalable GenAI solution powered by retrieval augmented generation (RAG) — unlocking 30–40% efficiency gains for analysts by transforming over 2 million customer reviews into actionable product insights.
Turning feedback into faster, smarter decisions
adidas has always led with innovation — starting with screw-in studs that transformed soccer cleats, and continuing through performance gear that blends style, sustainability and technology. To uphold this legacy and better serve their global customer base, adidas sought to tap into a massive, underused asset: over 2 million product reviews.
The goal? Use GenAI to extract sentiment and actionable insights from customer feedback, empowering more than 50 decision-makers globally, including nontechnical users, to improve product development and brand experience. A RAG chatbot would automate review analysis and deliver context-rich responses in real time, ensuring teams could act quickly on what customers actually wanted.
But adidas’ existing infrastructure couldn’t support that vision. Their legacy chatbot wasn’t built on GenAI or RAG, leading to generic answers, high compute costs and frustrating 15-second response times. Meanwhile, review analysis remained largely manual, which was time-consuming, and nontechnical users struggled to access insights. “With query payloads exceeding 200,000 tokens, we were overloading our back-end systems and limiting adoption,” Rahul Pandey, Senior Solutions Architect at adidas, said.
Building a scalable foundation for GenAI
adidas began their transformation by preparing a robust data layer to support RAG. First, over 2 million product reviews were embedded using models like Databricks BGE Large, optimized for semantic search. These vectorized reviews captured the context of customer feedback, not just keywords. The embeddings were then indexed using Mosaic AI Vector Search, Databricks’ native solution for fast, meaning-based retrieval.
With the vector database in place, adidas deployed a RAG pipeline using Model Serving. The pipeline retrieved relevant review snippets, combined them with user prompts and generated responses using LLMs such as Claude Haiku. Unity Catalog ensured secure access and governance of models and data, while supporting components like APIs and the chatbot interface were hosted in a cloud-native Kubernetes environment. A clean, intuitive front end built with React and Angular made it easy for teams across regions to access and interact with the tool.
To manage experiments and maintain system reliability, adidas used MLflow to track model performance and iterate efficiently. Feedback loops and error tracking ensured the chatbot stayed aligned with evolving customer sentiment, growing smarter with every interaction.
Consolidating 2 million reviews for deeper feedback analysis
The results were transformative. Latency dropped by 60%, cutting average response time from 15.5 seconds to six. More efficient prompt engineering and smaller context windows reduced token input size by 98.5%, from 200,000 to just 3,000 tokens per query. These optimizations lowered compute costs by over 90%.
Through GenAI, adidas improved review analysis efficiency by up to 30–40%, cutting down extensive manual workloads across global teams. The intuitive chatbot enabled faster decision-making across teams, from design and product to marketing and customer service. Nontechnical users could now extract insights independently, improving cross-functional collaboration and product development cycles.
“The GenAI infrastructure we built is already enabling use cases beyond product reviews,” Rahul said. “Teams are now exploring applications in customer service, knowledge management and other feedback channels.”
By integrating Databricks technologies like Mosaic AI Vector Search, Unity Catalog, Model Serving and MLflow, adidas created a global-scale solution that is fast, governed and extensible. It not only supports the company’s goal, but also sets the stage for a more personalized, responsive and data-driven future.