Generative AI Optimizes Incentivized Advertising Under Strict Constraints.

Incentivized advertising, which uses rewards to boost user engagement, faces significant challenges in optimizing continuous incentive levels while adhering to strict global budget or ROI constraints. Traditional methods struggle with high-frequency interactions, delayed feedback, and user fatigue. This new framework, GOAL (Generative Optimization for Incentivized Advertising with Global Level Constraints), addresses these issues by modeling incentive allocation as a conditional sequence generation task. GOAL generates incentive magnitudes based on user history and system-level constraints, employing a hierarchical causal state encoder to capture both short-term user behavior and long-term dependencies. It also introduces Safe Constrained Policy Optimization (SCPO), a method that allows a single generative policy to adapt to various ROI constraints without needing retraining. Experiments on real-world data and a simulated environment show that GOAL significantly improves long-term revenue and user retention. It also substantially reduces the rate of ROI constraint violations compared to existing baseline approaches, demonstrating its effectiveness in complex advertising scenarios.