Uplift modeling estimates the effect of a treatment on an individual’s outcome. However, accurately estimating this effect is challenging in multi-treatment scenarios due to data sparsity and class imbalance. While existing methods handle treatments independently, we introduce a novel paradigm that explicitly models and exploits local treatment similarities. To this end, we propose a non-parametric Bayesian criterion that automatically identifies and groups treatments with similar causal effects on specific subpopulations. We leverage this grouping strategy in two distinct ways: as an information pooling technique to improve state-of-the-art methods, and as a standalone uplift estimator that directly predicts treatment effects. Experimental results on diverse synthetic datasets demonstrate that our approach significantly improves the accuracy of uplift estimation compared to state-of-the-art methods.