Applying Causal Inference and Uplift Modeling for User Growth: Concepts, Methods, and Practice

The article begins with an overview of causal inference, explaining the difference between correlation (mere association) and causation (directional, necessary link), and highlights why correlation alone cannot determine treatment effects in user growth contexts.

It then outlines the three main streams of causal inference: computer science (Judea Pearl’s causal graph model and back‑door/front‑door criteria), econometrics (potential outcomes, double machine learning, DID, synthetic control, instrumental variables), and statistics (potential‑outcome framework, AB testing assumptions, and alternative methods).

Next, the focus shifts to uplift modeling for marketing, describing the classic Uplift (or causal) model taxonomy—Persuadables, Sure‑things, Lost‑causes, and Sleeping‑dogs—and illustrating how traditional response models can mislead coupon allocation, whereas uplift‑based decisions improve revenue.

The implementation section presents three practical algorithms: T‑learner (separate models for treatment and control), S‑learner (treatment as a feature in a single model), and a tree‑based uplift model that splits nodes to maximize uplift gain. Code snippets (importing LightGBM, data preprocessing, model training, and evaluation) are shown as images.

Model evaluation is discussed on two levels: effectiveness (using Qini curves and AUUC metrics to rank uplift scores) and business value (calculating uplift response rate and net incremental revenue). The article also covers more complex scenarios such as multiple coupon types, continuous treatments, and cost‑aware optimization for intelligent outbound call systems.

Finally, challenges are identified—including confounder identification, scenario‑specific adaptation, and scaling to large‑scale data—and future directions point to integrating causal inference with large language models and agents. Recommended reading includes the book "Causal Inference in Statistics" and several recent papers on multiple treatments, non‑randomized studies, real‑world applications, and uplift evaluation, with a GitHub repository offering additional code and resources.