FunnelCausalNet Optimizes Coupon Campaigns for Conversion and Revenue.

Coupon campaigns aim to boost both customer conversions and overall revenue. However, revenue, or Gross Merchandise Value (GMV), is inherently linked to conversion and subsequent order value, often exhibiting a zero-inflated and heavy-tailed distribution. This research proposes FunnelCausalNet, an uplift estimator that explicitly models this funnel by combining a binary conversion prediction head with a non-negative conditional value prediction head. The core idea is that by understanding the sequential nature of the customer journey, the model can more accurately predict the combined uplift in both conversion and revenue. Theoretical analysis suggests that this funnel-aware composition can reduce prediction variance under certain assumptions. The estimator is also equipped with tools for auditing and a Lagrangian allocator for budget-aware ROI optimization. Evaluations on semi-synthetic and real-world industrial datasets, including hotel coupon logs, demonstrated FunnelCausalNet's effectiveness. It achieved competitive or superior performance compared to eleven baselines, particularly in maximizing DeltaROI across various budget levels, and significantly reduced GMV effect error compared to direct GMV regression.