This paper proposes a Robust Causal Inference Framework (RCIF) for optimizing Direct Marketing Campaigns by moving beyond simple predictive modeling to accurately estimate the heterogeneous impact of treatment. Addressing critical econometric challenges such as selection bias and data leakage, the framework employs a multi-stage approach. In the first stage, we define the causal variables, strictly enforcing the pre-treatment assumption by identifying previous campaign contact as the Treatment (T) and eliminating leakage features such as call duration. The second stage establishes causal validity via Propensity Score Estimation. Using a Random Forest classifier, the model achieves an Area Under the Curve (AUC) of 0.6202, demonstrating good discriminative power without perfect separation. Crucially, the distribution analysis confirms sufficient overlap between treatment and control groups (overlap range: [0.050, 0.463]), thereby satisfying the Positivity Assumption required for stable causal inference. Campaign intensity and customer age emerge as the most important covariates in determining treatment assignment. The final stage utilizes Causal Meta-Learners (S-Learner, T-Learner, X-Learner, and Ensemble) to estimate the Conditional Average Treatment Effect (CATE), \(\tau (X)\) ). Segment-level analysis reveals positive treatment effects with a mean ATE of 0.082. Business simulation demonstrates that targeting customers selected by the S-Learner model yields a Return on Investment (ROI) improvement of +122.6% points over random targeting, with an additional profit of \(\$61,295\) and 102.5% more conversions. RCIF provides a validated, data-driven methodology for precision targeting and is recommended for deployment with A/B test validation.