Attentional Funnel Mapping: Deep Neural Attribution for Stage-Aware Marketing

• Ananya Sharma

  Author

Understanding the consumer's path to purchase is critical for effective digital advertising. Traditional models either predict conversions without explicit journey interpretation or assign funnel stages using methods lacking in transparency and scalability. This paper introduces a deep sequential architecture that learns to map raw user behavior trails to latent purchase funnel stages using novel global attention mechanisms. Our funnel-aware attention assigns interpretable, advertiser-specific importance scores to individual activities, enabling two core tasks: accurate conversion prediction and automatic, human-aligned behavior tagging. Evaluated on large-scale advertising data from three e-commerce advertisers, as well as the public RecSys15 dataset, the proposed neural attribution system outperforms several strong baselines, including multi-head self-attention and local attention models, achieving up to +7.04 percentage points absolute AUC improvement over a logistic regression baseline on proprietary data (corresponding to a relative lift of approximately 10%). Furthermore, the derived activity tags demonstrate close alignment with human expert annotations (RMSE = 0.96-1.18). In live production A/B tests, targeting users with creatives tailored to their inferred funnel stage drove a 3% to 6% incremental lift in conversion rate, while simultaneously improving click-through rates and reducing acquisition costs. These results validate that jointly modeling user journeys and interpretable funnel states within a single neural framework enhances both predictive performance and practical personalization in digital marketing.

• PDF

Copyright (c) 2026 International Journal of Adaptive Management and Business Intelligence

This work is licensed under a Creative Commons Attribution 4.0 International License.