Analyze control and variant performance with clear statistics. Review lift, confidence, power, and sample balance. Make better experiment calls using fast downloadable visual reports.
The chart compares observed conversion rates and the confidence interval of the rate difference.
| Variation | Visitors | Conversions | Conversion Rate | Revenue per Conversion | Estimated Revenue |
|---|---|---|---|---|---|
| Control A | 10,000 | 820 | 8.20% | $45 | $36,900 |
| Variant B | 9,800 | 910 | 9.29% | $45 | $40,950 |
Use rows like these when validating the calculator with your own experiment logs.
Conversion rate: p = conversions / visitors.
Pooled rate: p̂ = (c₁ + c₂) / (n₁ + n₂).
Pooled standard error: SE = √[p̂(1 − p̂)(1/n₁ + 1/n₂)].
Z score: z = (p₂ − p₁) / SE.
P value: derived from the standard normal distribution using one-tailed or two-tailed rules.
Confidence interval for the difference: (p₂ − p₁) ± zcritical × SEunpooled.
Lift: (p₂ − p₁) / p₁.
Required sample per variation: estimated from baseline rate, target uplift, alpha, and desired power using a two-proportion normal approximation.
It tests whether two conversion rates differ beyond random sampling noise. The result combines a z score, p value, confidence interval, and lift summary for practical interpretation.
Use a two-tailed test when you care about either improvement or decline. It is the safer default for most product, marketing, and experiment review workflows.
A one-tailed test can fit when you defined a single directional hypothesis before running the experiment. Do not switch after seeing the data.
Lift shows the relative percentage change from the control rate. It helps translate significance into business language, especially when comparing improvements across experiments.
Small samples can create large-looking differences that remain noisy. Significance depends on both effect size and uncertainty, not on lift alone.
Observed power gives an approximate sense of sensitivity for the current result. Use it carefully, and rely more heavily on planned sample size before launching tests.
The estimator uses a two-proportion normal approximation based on baseline rate, desired uplift, alpha, and target power. It returns the sample needed for each variation.
It is built for binary conversions. For average order value, revenue per user, or retention, use tests designed for continuous or time-based outcomes.
Important Note: All the Calculators listed in this site are for educational purpose only and we do not guarentee the accuracy of results. Please do consult with other sources as well.