Planned end date calculator for A/B tests
Find the date an A/B test should end from its baseline rate, minimum detectable effect, significance level, power, traffic split, and daily visitors, with a calendar showing the power reached on each day and the end date rounded up to whole weeks.
Planned end date
Power reaches 80% on Tuesday, 2 February 2027; the test should end on Sunday, 7 February 2027 so it covers whole weeks. Each variant needs 14,751 visitors, which takes 30 days of traffic at 1,000 visitors a day.
Power reached on each day
- Below 50%
- 50–79%
- 80% and up
| Mon | Tue | Wed | Thu | Fri | Sat | Sun |
|---|---|---|---|---|---|---|
4 January, 8% power | 5 January, 11% power | 6 January, 14% power | 7 January, 17% power | 8 January, 21% power | 9 January, 24% power | 10 January, 27% power |
11 January, 30% power | 12 January, 34% power | 13 January, 37% power | 14 January, 40% power | 15 January, 43% power | 16 January, 46% power | 17 January, 48% power |
18 January, 51% power | 19 January, 54% power | 20 January, 56% power | 21 January, 59% power | 22 January, 61% power | 23 January, 63% power | 24 January, 65% power |
25 January, 67% power | 26 January, 69% power | 27 January, 71% power | 28 January, 73% power | 29 January, 74% power | 30 January, 76% power | 31 January, 77% power |
| Mon | Tue | Wed | Thu | Fri | Sat | Sun |
|---|---|---|---|---|---|---|
1 February, 79% power | 2 February, 80% power | 3 February, 81% power | 4 February, 83% power | 5 February, 84% power | 6 February, 85% power | 7 February, 86% power, planned end date |
How the planned end date is calculated
The calculator first finds the visitors n that each variant needs, with the two-sided two-proportion z-test behind the sample size calculator, then divides n by the visitors a variant receives each day to count the days of traffic. Power is first reached on that day, and the planned end date is the Sunday that closes the Monday-to-Sunday week it falls in, so the result covers whole weeks.
days = ⌈n ÷ (daily visitors × share of traffic in each variant)⌉
Each day in the calendar shows the power of the visitors gathered through that day, which stays below the power you set until the sample is reached, and the test should be read once, at the planned end date, because stopping the first time it looks significant raises the false positive rate above the confidence level you chose.