方法对比

	Uplift Modeling	Online Controlled Experiment
领域	Marketing Science	Marketing Science
方法族	Machine learning	Process / pipeline
起源年份	2011	2020
提出者	Nicholas J. Radcliffe & Patrick D. Surry	Ron Kohavi, Diane Tang & Ya Xu (modern web experimentation practice)
类型	Heterogeneous-treatment-effect model for targeting incremental responders	Randomized-experiment pipeline for causal measurement of online changes
开创性文献	Radcliffe, N. J., & Surry, P. D. (2011). Real-World Uplift Modelling with Significance-Based Uplift Trees. Stochastic Solutions White Paper TR-2011-1.	Kohavi, R., Tang, D., & Xu, Y. (2020). Trustworthy Online Controlled Experiments: A Practical Guide to A/B Testing. Cambridge University Press. ISBN: 9781108724265
别名	Incremental Response Modeling, True-Lift Modeling, Net-Lift Modeling, Persuadable Targeting	A/B Testing, Split Testing, Randomized Web Experiment, Controlled Experiment on the Web
相关	3	3
摘要	Uplift modeling targets the people a marketing action actually changes, not the people most likely to buy anyway. Where a conventional response model predicts the probability of purchase, an uplift model predicts the difference a treatment makes — the incremental effect of, say, sending a coupon — and uses it to find 'persuadables' while avoiding 'sure things,' 'lost causes,' and especially 'sleeping dogs' who react negatively to contact. Nicholas Radcliffe and Patrick Surry, pioneers of the technique, formalized significance-based uplift trees that split on the difference in treatment-versus-control response rather than on response alone, and introduced the Qini curve to evaluate incremental gain. Pierre Gutierrez and Jean-Yves Gerardy's literature review situates uplift modeling squarely within causal inference, organizing the main estimation strategies and metrics. Because the quantity of interest is a conditional average treatment effect, uplift modeling is most reliable when built on randomized treatment and control data. The payoff is sharper, more profitable targeting: spend marketing effort where it produces genuine incremental response instead of rewarding behavior that would have happened regardless.	Online controlled experiments, commonly called A/B tests, randomly split live web or app traffic between a control and one or more treatment variants to measure the causal effect of a change on user behavior. Ron Kohavi, Diane Tang, and Ya Xu — who built and ran experimentation platforms at Microsoft, Google, and LinkedIn — set out the modern theory and best practice in their 2020 Cambridge book, and Kohavi's earlier survey with colleagues established the practical foundations of running trustworthy web experiments at scale. The discipline centers on a clearly defined Overall Evaluation Criterion (OEC) that captures long-term value, rigorous randomization, adequate statistical power, and a battery of trustworthiness checks such as the Sample Ratio Mismatch test. Because users are randomized, the difference in metrics between variants is an unbiased estimate of the change's causal impact — the gold standard for marketing and product decisions that attribution and observational analysis can only approximate. The output is a confident ship/no-ship decision: did this headline, layout, price, or feature actually move the metrics that matter, by how much, and with what certainty?