Budget-Constrained Causal Bandits: Bridging Uplift Modeling and Sequential Decision-Making

Budget-Constrained Causal Bandits (BCCB) unifies uplift modeling, exploration, and budget pacing into a single sequential framework, replacing the brittle two-stage offline HTE-then-optimize pipeline. Targeting cold-start ad campaigns where historical data is scarce, BCCB makes per-user treatment decisions online while simultaneously learning individual-level response heterogeneity. Practitioners in digital advertising running new campaigns or entering new markets should prioritize this over static constrained optimization approaches that require pre-collected data.