Dream11's AI Catalog Personalization Drives Revenue, User Satisfaction

Dream11, the fantasy sports platform, has successfully deployed a Hierarchical Contextual Uplift Bandit system for catalog personalization, resulting in a significant boost to both revenue and user satisfaction. This marks a major win for the application of advanced AI techniques in real-world business scenarios. The system, which went live in May 2025, leverages contextual similarity and uplift modeling to provide more relevant and personalized recommendations to users.

Addressing the Dynamic Nature of Fantasy Sports

Traditional Contextual Bandit (CB) algorithms often falter in dynamic environments. The fantasy sports landscape is characterized by rapid shifts in user behavior and reward distributions, driven by external factors like player performance, injuries, and team trades. This volatility necessitates frequent retraining of recommendation models, a process that can be computationally expensive and time-consuming. The Hierarchical Contextual Uplift Bandit framework addresses these challenges by dynamically adjusting contextual granularity. It intelligently shifts between broad, system-wide insights and detailed, user-specific contexts. This adaptive approach allows for effective policy transfer, mitigating cold-start issues and ensuring recommendations remain relevant even in the face of rapidly changing conditions.

The integration of uplift modeling principles is another key aspect of this system. Uplift modeling focuses on identifying users who are most likely to be positively influenced by a particular recommendation. By targeting these users with personalized suggestions, Dream11 can maximize the impact of its recommendations and drive incremental revenue. According to the arXiv pre-print, large-scale A/B testing prior to deployment demonstrated a 0.4% revenue improvement compared to the existing production system. Following the full production deployment in May 2025, Dream11 observed an additional 0.5% revenue increase, bringing the total revenue uplift to nearly 1%.

Implications for Personalized Recommendation Systems

The success of Dream11's Hierarchical Contextual Uplift Bandit system has significant implications for the broader field of personalized recommendation systems. It demonstrates the potential of AI to enhance user engagement and drive revenue growth in dynamic environments. While a 1% revenue improvement might seem marginal, for a company the size of Dream11 the impact on the bottom line is substantial. Furthermore, the reported improvements in user satisfaction metrics suggest that the system is not only driving revenue but also enhancing the overall user experience.

This approach offers a robust framework for adapting to evolving user preferences, a capability that is increasingly critical in today's fast-paced digital landscape. Other companies with large catalogs and dynamic user bases would be wise to take note. The framework’s ability to dynamically adjust to changing conditions, coupled with its focus on uplift modeling, provides a powerful toolkit for optimizing recommendation strategies. It's a significant step forward from traditional CB algorithms, and it is likely we will see broader adoption of similar techniques across various industries in the coming years, especially as the cost of computation continues to decrease.