Comparer des méthodes

	Test A/B pilote	Test A/B adaptatif
Domaine	Plans d'expériences	Plans d'expériences
Famille	Process / pipeline	Process / pipeline
Année d'origine	2000s–2010s (formalized in digital experimentation literature)	1952 (Robbins); applied to A/B testing from ~2010s onward
Auteur d'origine	Derived from pilot study methodology (Kraemer et al., 2006) applied to A/B testing practice	Herbert Robbins (bandit framework); Thompson Sampling formalized by William R. Thompson
Type	Experimental design — feasibility study	Adaptive experimental design
Source fondatrice	Thabane, L., Ma, J., Chu, R., Cheng, J., Ismaila, A., Rios, L. P., Robson, R., Thabane, M., Giangregorio, L., & Goldsmith, C. H. (2010). A tutorial on pilot studies: The what, why and how. BMC Medical Research Methodology, 10(1), 1.	Russo, D., Van Roy, B., Kazerouni, A., Osband, I., & Wen, Z. (2018). A Tutorial on Thompson Sampling. Foundations and Trends in Machine Learning, 11(1), 1–96.
Alias	pilot split test, feasibility A/B test, preliminary A/B experiment, pilot randomized comparison	adaptive AB test, bandit A/B test, multi-armed bandit testing, online adaptive experiment
Apparentées	5	6
Résumé	A Pilot A/B test is a small-scale, preliminary split-test experiment run before a full A/B test to assess feasibility, estimate effect sizes, detect operational problems, and validate measurement instruments. Participants are randomly assigned to a control condition (A) and a treatment condition (B), but the study is explicitly underpowered — its purpose is to inform the design of the definitive test, not to yield a conclusive comparison.	An Adaptive A/B test is an experimental design that dynamically reallocates traffic or participants toward better-performing variants during the experiment itself, rather than holding allocations fixed until the end. Drawing on multi-armed bandit algorithms such as Thompson Sampling or Upper Confidence Bound (UCB), it balances the exploration of uncertain variants with the exploitation of those already showing superior performance, typically yielding higher aggregate outcomes while still producing valid inferential conclusions.