Skip to main content

COMPEL Glossary / a-b-test-rct

A/B test (RCT)

A randomised controlled trial in which units are randomly assigned to treatment (AI feature) and control (no feature or baseline feature).

What this means in practice

Gold-standard causal-inference design because randomisation breaks confounding; limitations include contamination, novelty effects, and external validity when piloting in a narrow subset.

Synonyms

randomised controlled trial , AB test for AI

See also

  • Difference-in-differences (DiD) — A quasi-experimental design comparing treated and control trajectories over time — identifying causal effect from differential change.
  • Regression discontinuity (RDD) — A quasi-experimental design using a threshold — e.g., a credit-score cutoff, an eligibility cutoff — to create a natural experiment near the cutoff.
  • Counterfactual outcome — The outcome that would have occurred without the AI intervention — the benchmark against which incremental AI value is measured.