Uncertainty-Aware Marketing Attribution Inference and Budget Decision-Making with Intelligent Agents

This paper addresses the issues of insufficient credibility in attribution inference and susceptibility to noise-induced fluctuations in budget allocation in multi-touchpoint marketing scenarios. It proposes a unified framework for marketing attribution inference and budget decision-making agents that incorporates uncertainty modeling. The method uses user interaction paths as sequence input, generating touchpoint weights through sequence encoding and importance modeling. Simultaneously, it outputs the expected incremental contribution and uncertainty characterization at the channel level, extending attribution results from single-point estimation to distributed signals usable for risk measurement. At the decision-making end, a risk-aware budget optimization objective is constructed, coupling contribution expectation and uncertainty penalty into the budget allocation process. Smoothing constraints are introduced to suppress frequent adjustments, forming a closed-loop update mechanism from data to attribution to budget, enabling the strategy to achieve a balance between revenue and stability under constraints. Multi-touchpoint path and cost characteristics are constructed based on publicly available programmatic advertising datasets. An evaluation system covering attribution error, probability calibration, and budget stability is designed. Comparative experiments verify the framework's comprehensive advantages in attribution reliability and budget decision quality, demonstrating the crucial role and engineering usability of uncertainty in the attribution-to-decision transmission process.