Talk
Most real-world audience modeling systems rely on first-party behavioral data, which captures what users do but not why they do it. This limitation often results in segments that are statistically consistent but lack interpretability and strategic relevance.
This talk presents a hybrid framework that combines survey-driven persona discovery with machine learning-based activation to address this gap. First, unsupervised learning techniques are applied to survey data to identify intent-driven personas grounded in motivations and preferences. These personas are then mapped to large-scale behavioral datasets using supervised learning models, enabling scalable prediction of persona membership.
A key architectural principle is the decoupling of persona discovery from activation. This allows qualitative insights to inform segmentation while maintaining scalability, interpretability, and privacy compliance. The talk will cover practical aspects of implementing this system, including feature alignment between survey and behavioral data, handling sparsity and bias in survey inputs, and trade-offs between interpretability and predictive performance.
The session will also highlight lessons learned from real-world deployments and provide a practical blueprint for building production-ready audience modeling systems that go beyond purely signal-driven clustering.
About the Speaker
Nikhil Singh is a Data Science Lead with over 12 years of experience in applied machine learning, experimentation, and large-scale analytics. He specializes in building scalable, privacy-aware AI systems that translate complex consumer behavior into actionable business insights. His work spans audience intelligence, predictive modeling, and the design of data-driven decision systems across high-scale environments.
He has led initiatives focused on persona modeling, customer segmentation, and measurement frameworks, with a strong emphasis on bridging the gap between observed behavior and underlying consumer intent. His recent work explores intent-aware AI systems that combine survey-based insights with machine learning to build scalable predictive audiences.
Nikhil was a guest speaker at the IEEE New Era AI World Leaders Summit (Seattle) 2025 and his work has also been accepted for presentation at the IEEE Swiss Conference on Data Science and AI (SDS 2026). He is particularly interested in designing interpretable and production-ready AI systems that balance scalability, privacy, and real-world applicability.