# A/B testing · Glossary
A controlled experiment in which users or cases are randomly assigned to two or more versions of something, such as a web page, message or process, and the outcomes are compared.
[Glossary](/glossary) · Analytics engineering and analytics

# A/B testing

     A controlled experiment in which users or cases are randomly assigned to two or more versions of something, such as a web page, message or process, and the outcomes are compared. Randomisation lets differences be attributed to the change rather than to other factors.

## Why it matters

     A/B testing gives stronger evidence of cause and effect than before-and-after comparisons, which are easily confused by seasonality and other changes. It is widely used to evaluate product features, pricing presentation and model-driven interventions.

     Reliable tests need a primary metric chosen in advance, enough sample size, a fixed duration and guardrail metrics that catch harm, such as increased complaints or returns.

## In practice

     For example, a UK online insurer might test a redesigned quote page against the current version, with conversion as the primary metric and cancellations within the cooling-off period as a guardrail.

## Where Rodan fits

     Rodan designs experiments and measurement for digital products and AI interventions in [Analytics and Intelligence](https://rodan.io/what-we-build/analytics-intelligence) work.

## Related terms

- [Data science](/glossary#data-science)

- [AI evaluation (evals)](/glossary#ai-evaluation)

- [Dashboard](/glossary/dashboard)

- [Prescriptive analytics](/glossary/prescriptive-analytics)
HTML: https://rodan.io/glossary/ab-testing