Relative: target = baseline × (1 + MDE)
Sample size 8,158 for control, 8,158 per variation, 16,316 in total, about 17 days.
Sample size
- Control
- 8,158
- Variation
- 8,158
- Total
- 16,316
- Estimated duration
- 17 days
Detects 5% → 6% (+1 pp) at 95% confidence and 80% power, two-sided.
Excludes ramp-up, weekly cycles and slow conversions. Run at least one full week.
Duration by MDE
Same inputs, different minimum detectable effects.
| MDE | Target rate | Control | Variation | Total | Days |
|---|---|---|---|---|---|
| 5% | 5.25% | 122,124 | 122,124 | 244,248 | 245 |
| 10% | 5.5% | 31,234 | 31,234 | 62,468 | 63 |
| 15% | 5.75% | 14,193 | 14,193 | 28,386 | 29 |
| 20%current | 6% | 8,158 | 8,158 | 16,316 | 17 |
| 25% | 6.25% | 5,333 | 5,333 | 10,666 | 11 |
| 30% | 6.5% | 3,780 | 3,780 | 7,560 | 8 |
| 40% | 7% | 2,213 | 2,213 | 4,426 | 5 |
| 50% | 7.5% | 1,471 | 1,471 | 2,942 | 3 |
Everything runs in your browser. The numbers you enter are never uploaded or stored.
How do I use the A/B test calculator?
Use Pre-Test Analysis before the test to find out how many users or sessions you need and how many days that takes. Use Test Analysis when the test has finished: enter the users or sessions and conversions for control and each variation, and the results update as you type.
- Choose Pre-Test Analysis, enter your baseline conversion rate, the smallest lift worth detecting (MDE), confidence, power and daily traffic.
- Read the sample size per arm and the estimated duration, and use the duration table to see how a bigger or smaller MDE changes it.
- Run the test for the planned duration without stopping early when the numbers look good.
- Choose Test Analysis, enter the test duration, then users or sessions and conversions for control and every variation.
- Read lift, p-value and confidence interval per variation, check the SRM result, then copy the summary or download a .csv.
Which statistical methods does the calculator use?
Frequentist, two-sided, fixed-horizon methods for a binary conversion metric. One conversion counts once per user or session.
- Sample size
- Normal approximation for two independent proportions, with the target rate = baseline × (1 + MDE). Uneven splits use the ratio between variation and control traffic.
- Significance
- Pooled two-proportion z-test, two-sided. A variation is significant when its p-value is below α.
- Confidence interval
- Newcombe hybrid Wilson-score interval for the difference in conversion rates, in percentage points.
- Several variations
- Bonferroni correction: with m variations, each comparison uses α ÷ m and its interval uses confidence 1 − α ÷ m, both in planning and in analysis.
- Sample ratio mismatch
- Pearson chi-square goodness-of-fit test of users per arm against the planned split. Flagged when p < 0.001.
- Additional days needed
- The sample needed to detect the observed difference at your confidence and power, minus what you have, at the daily traffic so far. It is an estimate that assumes the observed difference is the true one.
- Monthly impact
- Daily users or sessions across all arms × 30 × the difference in conversion rate, times the average value per conversion. A projection, not a forecast: the true difference can be anywhere in the confidence interval.
Not covered: Bayesian or sequential testing, revenue or other continuous metrics, several metrics at once, and early stopping. Checking results again and again and stopping when they look significant raises the false-positive rate, so decide the sample size first and analyze once.
Frequently asked questions
Is the MDE relative or absolute?
Relative. A 20% MDE on a 5% baseline means detecting a change from 5% to 6%, an absolute difference of 1 percentage point. The calculator shows both the target rate and the absolute difference so there is no doubt.
Should I enter users or sessions?
Use the unit your test randomizes on, and count each one once. If users are assigned to variations, enter users and the users who converted. Mixing units, such as sessions with user-level conversions, makes the results unreliable.
Why does a test with several variations need more traffic?
Each extra variation is one more chance of a false positive. The calculator applies the Bonferroni correction, which tests each variation at a stricter level, so you need more users per arm to detect the same lift.
What does a sample ratio mismatch (SRM) mean?
The split of users between arms is too far from what you planned to be chance. It usually points to a problem with assignment, redirects, bots or tracking. It is a warning to investigate, and results with SRM should not be trusted until it is explained.
Is my data sent anywhere?
No. Every calculation runs in your browser and the numbers you enter are never uploaded or stored. The site only records anonymous usage, such as a calculation being run, without any of your figures.
Is the A/B test calculator free?
Yes. It is free, needs no account or sign-up and has no limits on the number of calculations.
Who built this tool?
I'm Reinaldo Nani, a MarTech Stack Architect & DXP Manager based in São Paulo, Brazil. I run CRO programs and A/B tests for enterprise clients, and this is the calculator I use to plan them and to read their results.