Uber Eats reduces search latency by 50% through system optimization

Uber Eats has successfully reduced its end-to-end search latency by 50% through a comprehensive overhaul of its core technical components. The optimization efforts spanned the entire processing chain, including retrieval, feature hydration, ranking, advertising delivery, and IT infrastructure.

Key technical improvements included the adoption of an ‘Above-the-Fold-Completion’ metric, which improved performance by 200 ms by focusing on the completion of the immediately visible screen area. The company also streamlined data retrieval by removing low-value retrieval strategies, saving 120 ms, and implemented product-based embeddings that reduced necessary data queries by more than 100-fold.

Additionally, Uber Eats restructured its data architecture by decoupling ranking hydration from presentation data. This separation ensures that information intended solely for visual UI rendering does not need to pass through computationally intensive evaluation stages, resulting in a saving of over 100 ms.