See how many visitors your A/B test needs.
Enter your current conversion rate and the smallest change worth testing. Add daily traffic to estimate how long your test might take.
Your test inputs
The signup rate on your current page, before you make a change.
Count visitors who could see either version of the page.
Sample size
for each version
About 2,060 visitors in total, split evenly between both versions.
How long could this take?
Enter your daily visitors to see roughly how long this could take.
Include a normal week of traffic, even if you reach the visitor target sooner. Longer buying cycles may need more time.
This plan looks for a change of 5 percentage points. Smaller changes can still happen, but they are harder to confirm.
How long should an A/B test run?
Start with the total visitors needed for both versions, then divide by the visitors who could enter the test each day. At the default settings, you need 2,060 visitors in all. With 500 eligible visitors a day, that is about 5 days to collect the sample.
Five days is an estimate, not a signal to stop. Keep the test running long enough to include your usual weekday and weekend pattern. This estimate assumes steady daily traffic and an even split between the two versions.
Choose the inputs before you start
Decide which result you want to improve, such as signups, and what change you think could help. Use that result to plan and judge your test.
Baseline conversion rate
The percentage of eligible visitors who complete your chosen action before the test. If 20 out of 100 visitors sign up on the current page, its baseline signup rate is 20%.
Minimum detectable effect
The smallest change you want a good chance of finding. If signups are at 20% and a rise to 25% would matter, choose an absolute change of 5 percentage points. This is a planning target, not a pass-or-fail rule for the result.
Power and significance
Power is your chance of finding the change you planned for if it is real. At 80% power, a real rise from 20% to 25% would show up in about 8 of 10 similar tests. A 5% significance level means about 5 in 100 tests with no real change could still look like a win.
Keep a test log. Write down how many visitors each version needs. When the test ends, record each version's signups and signup rate, your decision, and what you learned. Adam Treboutat explains this in the webinar recording.
The sample size sets the plan. The results decide the winner.
Compare the conversion rates after the planned run. A version wins when its improvement passes the statistical check you chose before the test and is worth acting on. Otherwise, there is no clear winner.
A/B test sample size questions
How do I calculate A/B test sample size?
Enter the conversion rate you have now and the smallest change worth testing. The calculator estimates how many visitors each version needs. With the default settings, that is 1,030 visitors per version, or 2,060 total. It assumes two equally sized groups and a yes-or-no result, such as whether someone signs up.
What if I do not get enough visitors?
Add your daily visitors to see roughly how long the test could take. Planning for a bigger change means you need fewer visitors, but you are less likely to spot a smaller improvement. If it would still take too long, try a higher-traffic page or another way to learn. Decide what change matters before you see the results.
Can I stop when I have enough visitors?
Not automatically. First check that both versions got a fair split and conversions were counted correctly. After the planned number of visitors and a normal traffic cycle, compare their conversion rates in your testing tool. A version wins when its improvement passes the statistical check you chose before launch and is worth acting on. If the evidence is unclear, call the test inconclusive. Do not stop early just because one version looks ahead on a daily check.
Does this work for revenue or more than two versions?
No. This calculator compares two versions that receive the same number of visitors. It counts a yes-or-no result, such as a signup. It does not work for average revenue, three or more versions, or an uneven split.
How this number is calculated
This estimate uses the same math as Evan Miller's calculator. It compares two versions using a yes-or-no result, such as a signup, and gives you the number of visitors needed for each version.
A smaller change is harder to spot and usually needs more visitors. Asking for a better chance of spotting it, or fewer false alarms, also needs more visitors.
This tool helps you plan a test. It does not analyze results or adjust for repeated checks, more than two versions, or an uneven split.