Modelling Approach for enhanced market expansion in Digital Marketing in B2B space

Abstract

This research delves into lookalike modelling, a data-driven strategy to identify potential customers with characteristics similar to high-value customers, thereby expanding market reach. The study constructs a seed set using job-related attributes, employing machine learning to discern commonalities among individuals with elevated lifetime value or pronounced brand engagement. A fast multi-graph-based approach recommends lookalike users from a vast dataset, demonstrating computational efficiency. A distributed architecture ensures scalability, achieving a consistent recommendation throughput of 500 recommendations per second.

Performance evaluation on real-world data shows a substantial improvement in precision performance, reaching up to 90% for 200 lookalike candidates in digital marketing. The research introduces a comprehensive preprocessing pipeline for diverse seed set data, addressing noise and volume issues and enhancing downstream model generation applicability. The study also promotes collaboration and transparency through an open-source module for replicability in similar projects.