Estimate an approximate per-variant sample for a Bayesian conversion experiment using a target posterior probability, minimum detectable relative lift, and optional prior strength. The approximation translates the posterior threshold into a normal probability target and shows how prior information affects new-data needs.
The estimate is a planning approximation for the sample needed to make the probability of a positive difference reach the chosen threshold at the assumed effect.
Bayesian stopping behavior depends on the full prior, decision rule, and predictive design. Validate important experiments with simulation.
For a 10% baseline, 10% relative lift, 95% posterior target, and no prior effective sample, the normal approximation requires about 5,084 new users per variant.
It uses the inputs, formula, units, and independence assumptions shown on this page. Change the inputs to match your own system or experiment.
Use representative averages for planning, preserve headroom where applicable, and test multiple plausible scenarios instead of relying on one point estimate.
No. The result is a planning estimate. Validate it with measured data, monitoring, load tests, or an experiment-design review as appropriate.
The calculator checks values required by the formula and displays an error instead of returning NaN or Infinity. Valid zero values remain available where they are meaningful.
Tools may use different statistical approximations, unit conventions, rounding rules, priors, confidence definitions, or operational headroom assumptions.
| Variable | Unit | Meaning |
|---|---|---|
| Posterior target | probability | Required probability that variant exceeds control |
| Prior effective n | users/variant | Information represented by the prior |
| New n | users/variant | Approximate additional observations |