A/B Test Duration Calculator

This is the time tool. It answers "how many days and when it ends", ideal for fitting a test into the calendar and deciding whether it fits. If what you want is the number of visitors per variation, use the sample size calculator. Both share the same statistical base, they change the angle: here the main output is the runtime and the date.

How to use it

1. Enter your current conversion rate and the minimum detectable effect (the smallest gain that changes your decision).
2. Pick the number of variations, counting the control (2 is a simple A/B).
3. Enter your total weekly traffic on the tested page.
4. Read the days, the total visitors and the projected finish date, plus the note on whether the timeline is healthy or too long.

How it works: the formula

First the calculator sizes the sample per variation with the two-proportion normal approximation. Then it divides the total by daily traffic:

The n per variation comes from the same math as the sample size calculator. The total multiplies by how many variations there are, and weekly traffic becomes daily traffic to reach the days. The finish date is today plus those days, in your timezone.

Worked example (reproduces the default result)

With the pre-filled values: rate 5%, MDE +10% relative, 95% confidence, 80% power, 2 variations and 20,000 visitors per week. The sample is 31,234 per variation, or 62,468 total. Daily traffic is 20,000 / 7 = 2,857. So days = ⌈62,468 / 2,857⌉ = 22 days, about 3 weeks. That is what the tool shows above.

How to read it, and where it fools you

Treat the duration as a floor, not an exact ceiling. It assumes stable traffic and an even split across variations. In practice, always close whole weekly cycles: if it says 22 days, plan 3 full weeks so you do not end mid-week on an odd Monday.

If the timeline comes out too long, the problem is almost never the tool, it is the test. Raise the MDE, drive more traffic, or cut variations. And remember: running to the date is worthless if you peek and stop early, that is the peeking problem. When it ends, check the result with the significance calculator.

Best practices when planning the runtime

The duration is only worth something if you respect the calendar it suggests. Before you schedule the test, line up these points with whoever owns the page.

• Always round up to whole weeks. A test that ends on a Wednesday mixes partial weeks and skews the reading.
• Reserve stable traffic for the window. A media campaign mid-test changes the audience and distorts the result.
• If the runtime runs past a few weeks, prefer fewer variations. A simple A/B answers far faster than an A/B/C.
• Write down the finish date and treat it as a commitment. Ending early out of impatience is the most common way to burn a well-planned test.

Frequently asked questions

How long does an A/B test need to run?

Long enough to gather the required sample across all variations, without stopping early. It depends on your conversion rate, the effect you want to detect, the number of variations and your traffic. For a 5% rate, +10% relative and 20,000 visitors per week across two versions, that is about 22 days.

How is this different from the sample size calculator?

The sample size calculator answers "how many visitors per variation". This one answers "how many days and when it ends", already accounting for the number of variations and your traffic. Use the sample one to size, this one to plan the test calendar.

Should I run the test for at least one week?

Yes, always close whole weekly cycles. Buying behavior differs between weekdays and weekends, so finishing mid-week distorts the result. If the duration comes out at 9 days, round up to 14.

What do I do when the duration is too long?

Three levers: raise the minimum effect you accept detecting (a larger MDE shortens the test a lot), drive more traffic to the tested page, or cut the number of variations. Testing fewer things at once is almost always the fastest path to an answer.

Do more variations make the test slower?

Yes. Traffic splits across all groups, so each one takes longer to reach its sample. A well-sized A/B/C test takes well over twice as long as a simple A/B on the same traffic.

Keep going

To understand why time and sample move together, see how to run an A/B test and the guide on what A/B testing is.