How Adidas Analyzes 2 Million Reviews 40% Faster with Databricks GenAI

Adidas has built its legacy on innovation—from screw-in studs that changed soccer to performance gear that blends style, sustainability, and technology. Serving customers in 150+ countries, the brand needed a faster way to understand what those customers actually wanted by analyzing product feedback at scale. Over 2 million reviews existed across the product catalogue, but the infrastructure to make them useful didn't.

The legacy chatbot system couldn't support a GenAI vision. Query payloads exceeded 200,000 tokens per request, overloading back-end systems and limiting adoption. Response times averaged 15.5 seconds. Review analysis remained largely manual, time-consuming, and inaccessible to nontechnical users. "With query payloads exceeding 200,000 tokens, we were overloading our back-end systems and limiting adoption," said Rahul Pandey, Senior Solutions Architect at Adidas.

Adidas built a new foundation on Databricks. The 2 million product reviews were embedded using Databricks BGE Large, optimized for semantic search, and indexed using Databricks Vector Search for fast, meaning-based retrieval. A RAG pipeline using Databricks Model Serving combined retrieved review snippets with user prompts, generating responses with Claude Haiku as the primary LLM. Unity Catalog enforced secure data access and model governance. MLflow tracked model performance and experiment iteration. A clean React and Angular front end made the tool accessible to teams across regions and roles, including nontechnical users.

The results were immediate and precise. Latency dropped 60%, from 15.5 seconds to 6 seconds. Token input size fell 98.5%, from 200,000 to 3,000 per query, cutting compute costs by over 90%. Analyst efficiency in review-based decision-making improved 30-40%. The intuitive interface enabled faster cross-functional decisions across design, product, marketing, and customer service—and for the first time, nontechnical users could extract insights independently. "The GenAI infrastructure we built is already enabling use cases beyond product reviews," Pandey said. "Teams are now exploring applications in customer service, knowledge management and other feedback channels."

The platform was built to be extensible. By integrating Databricks Vector Search, Unity Catalog, Model Serving, MLflow, and Delta Lake, Adidas created a governed, global-scale solution that supports the brand's ambition for a more personalized and data-driven customer experience—and lays the infrastructure for whatever feedback channels come next.