Modeling Time-Sensitive Causal Uplift in Marketing Campaigns: Evidence From Synthetic and Real-World Data

Uplift modeling (or incremental response modeling) is a technique used in many business applications to estimate the causal impact of an intervention. In the potential outcomes approach, it has established a conceptual basis for marketing analytics and causal inference studies, enabling more effective targeting of customers whose behavior is affected. Rather than focusing only on who is likely to respond, it looks at why and when responses occur. The approach estimates individualized treatment effects (ITEs) to identify which customers react specifically because of a marketing intervention. The analysis proceeds in two stages. First, a controlled synthetic dataset is designed to simulate real-world variation in demographics, behavior, and treatment assignment bias. This dataset allows performance benchmarking through T-Learner models trained with gradient boosting classifiers. The second stage adapts the Hillstrom Email Analytics dataset to include elements of timing, using customer recency as an indicator of behavioral time. By dividing the data into time bins, the study explores how uplift and the area under the uplift curve (AUUC) change across engagement periods. Results show that treatment effects vary over time, with response gains peaking before gradually falling due to customer fatigue. Discount-based campaigns deliver the highest additional conversions in the synthetic tests, while in the Hillstrom data, recent customers demonstrate stronger causal lift than long-inactive ones. Overall, the framework brings together simulated precision and empirical evidence, showing how time-aware uplift modeling connects predictive analytics with more adaptive, ethically informed marketing decisions.