A/B Testing Memory Requirement Calculator FAQ

It means the visitor sample required to evaluate a planned conversion-rate change. It does not refer to RAM, storage, or website performance memory.

It estimates visitors needed per version, total eligible visitors, expected conversions, and the approximate number of days needed to collect the traffic.

Sample size is typically planned before an experiment starts so the team can assess traffic feasibility and define a target collection period.

Yes. Enter the total number of versions, including the control. The calculator multiplies the per-version requirement by that number.

It is the current or expected conversion rate for the control experience among eligible visitors.

It is the smallest relative improvement that would be meaningful for the test to detect. The calculator turns it into an absolute conversion-rate difference.

A change from 5.0% to 5.5% is a 10% relative uplift, so enter 10 as the minimum detectable uplift.

The available choices use standard critical values: 80%, 90%, 95%, or 99% confidence. Higher settings increase the sample estimate.

Power is the planned probability of detecting the chosen uplift if that uplift is truly present. Higher power generally requires more visitors.

No. It is the requirement for one version. Multiply it by the number of versions to obtain the total planned eligible traffic.

The calculator divides total visitors needed by daily eligible visitors and rounds up to the next whole day.

They are an approximate total number of conversions over the planned sample, using the average of the baseline and expected variant conversion rates.

Eligible traffic can vary, traffic allocation may not remain even, and operational constraints or a need to observe normal cycles can extend the calendar duration.

Use visitors who can actually enter the experiment and reach the relevant conversion opportunity, rather than unrelated site traffic.

No. It is a statistical planning estimate and depends on stable measurement, suitable randomization, a clear metric, and the assumptions used.

No. Tests with several variations, multiple primary metrics, or repeated checks may need a larger sample or a different analysis plan.

An early favorable result can be misleading if the stopping rule was not planned. Define how the experiment will be evaluated before launch.

It reduces false-positive risk in the calculation but also increases the traffic requirement. The appropriate setting depends on the experiment's planning context.

The displayed formula is designed for conversion rates, which are binary outcomes. Continuous or highly variable metrics usually need a different sample-size method.