Use our free A/B test results calculator to compare conversion rates, calculate conversion lift and assess whether an observed difference is statistically significant. Enter the visitor and conversion counts for your control and variation to see confidence intervals and a clear explanation of your results.
The built-in Sample Ratio Mismatch (SRM) checker also compares your actual visitor allocation with your planned traffic split, helping you identify potential data issues before interpreting the outcome.
What do your A/B test results tell you?
A higher conversion rate does not automatically mean your variation is a winner. Differences can occur through random variation, particularly when conversion counts are low. This calculator helps you understand the size of the observed change and the statistical evidence behind it.
Use it to:
-
Compare control and variation conversion rates.
-
Measure relative lift and the absolute difference in percentage points.
-
Assess statistical significance at a 90%, 95% or 99% confidence level.
-
Understand uncertainty through an approximate confidence interval.
-
Check whether visitor allocation is consistent with your planned traffic split.
How to use the calculator
Enter the unique visitors and converting visitors for each version over the same test period, counting each converting visitor once. Set your planned traffic allocation and confidence level, then select Calculate results.
Review the traffic check before interpreting conversion performance. Assess results at your planned stopping point, and consider commercial value and guardrail metrics alongside statistical significance.
Still planning your experiment? Use our A/B Test Duration & Sample Size Calculator to estimate the traffic and runtime needed before you launch.
A/B Test Results Calculator
Understand your conversion lift, assess the evidence and check your traffic split.
1. Enter your data
Use unique visitors assigned to each version over the same period. Count each converting visitor once.
Two independent groups and one preselected conversion metric. Higher conversion is treated as better.
2. Results
Enter your experiment counts, or load the example to see how the calculator works.
| Two-sided p-value | |
|---|---|
| Significance threshold (α) |
Traffic split check (SRM)
| Group | Expected visitors | Observed |
|---|---|---|
| Control | ||
| Variation |
Assess results at your planned stopping point. Repeated checking and stopping when significant increases false positives.
A statistical result does not confirm commercial value. Review guardrail metrics and implementation costs before rollout.
How to interpret these results
- Relative lift: the percentage change in conversion rate compared with control. A move from 3% to 3.6% is a 20% lift and a 0.6 percentage-point increase.
- Confidence interval: an approximate range of plausible absolute differences under the model. A range spanning zero includes both improvement and decline. An inconclusive result does not establish that the versions are equivalent.
- P-value: how unusual a difference at least this large would be if the true conversion rates were equal, under the test assumptions. It is not the probability that the variation is a winner.
- Traffic check: compares visitor allocation with your planned split. A warning calls for investigation; a pass does not prove that tracking is correct. Use original assignment counts, not downstream funnel counts.
Method, assumptions & limitations
Conversion inference uses a fixed-horizon, two-sided pooled two-proportion z-test. Absolute-difference intervals use an unpooled normal approximation, clipped to the possible −100 to +100 percentage-point range. The test and interval use different standard errors, so borderline decisions can differ slightly. Confidence choices use z = 1.644853627, 1.959963985 or 2.575829304.
Conversion inference is withheld unless each group has at least 10 observed conversions and 10 non-conversions, and each pooled expected success/failure count is at least 10. Rates and lift remain descriptive. Zero control conversions means relative lift is undefined.
The SRM check uses Pearson’s chi-square goodness-of-fit test with one degree of freedom and a fixed warning threshold of p < 0.001. Expected visitor counts must each be at least five. A detected or unavailable SRM check blocks an actionable conversion verdict.
Assumes random assignment, independent visitors, stable allocation and complete follow-up. Aggregate repeated visits to the assignment unit. Do not use raw event counts, revenue, average order value or overlapping audiences. Planned split changes require separate checks by allocation period. No sequential-testing or multiple-comparison adjustment is applied; do not use this tool to select the best of several variants.
Methods: NIST: difference of proportions · NIST: chi-square goodness of fit · Microsoft: diagnosing SRM