Iterative Causal Segmentation: Filling the Gap between Market Segmentation and Marketing Strategy

TL;DR

The paper tackles the gap between market segmentation and causal marketing strategy by addressing the tightly coupled relationship between segmentation and causal uplift. It introduces Iterative Causal Segmentation, a joint convergence framework that alternates between estimating uplift via causal methods and updating segmentation, with convergence defined by minimal segment movement relative to ATE-driven variance. Key formulations include the Average Treatment Effect ($ATE = E(Y^{A=1}) - E(Y^{A=0})$) and the Conditional Average Treatment Effect ($\text{CATE}(X) = E(Y^{A=1}|X) - E(Y^{A=0}|X)$), and a causal graph that guides confounder control. The method is validated on Uber CausalML data, demonstrating that segmentation driven by CATE yields superior uplift performance compared to propensity-score or KMeans baselines, and providing explainability through SHAP alongside sensitivity analyses via Qini curves. The work offers a principled, exclusivity-backed approach with practical implications for targeted marketing and uplift optimization in tightly coupled segmentation settings.

Abstract

The field of causal Machine Learning (ML) has made significant strides in recent years. Notable breakthroughs include methods such as meta learners (arXiv:1706.03461v6) and heterogeneous doubly robust estimators (arXiv:2004.14497) introduced in the last five years. Despite these advancements, the field still faces challenges, particularly in managing tightly coupled systems where both the causal treatment variable and a confounding covariate must serve as key decision-making indicators. This scenario is common in applications of causal ML for marketing, such as marketing segmentation and incremental marketing uplift. In this work, we present our formally proven algorithm, iterative causal segmentation, to address this issue.

Figures (9)

• Figure 1: Potential causal graph, where promotion is $A$, covariates are $X_i$, promotion outcome is $Y$, and marketing segments $S$ is a confounder.
• Figure 2: The diagram description on how to use causal machine learning to guide segmentation efforts. The goal is to achieve the joint convergence of the causal machine learning module as well as the convergence of the segmentation module.
• Figure 3: Segmentation results and sensitivity study result for the converged system described in Table \ref{['t:convergence']}
• Figure 4: Simulation study comparing the performance among the cumulative return gain of four different promotion population selection strategies. Causal Effect selection is based on the treatment effect. Propensity score selection is based on the propensity, which is regressed from the given $X$ to the propensity of obtaining outcome $Y$. KMeans is based on clustering results using $X$. Random selection is a random promotion assignment, whose expected slope is the same as the expected gain, i.e., ATE.
• Figure 5: Simulation study comparing the performance in terms of cumulative return gain among four different promotion population selection strategies. Specifically, the causal effect curve is labeled.

Theorems & Definitions (1)

• proof : Proof of Theorem 1