Free tool A/B test duration calculator
Turn a sample size into a calendar. Enter your traffic, the share of it that enters the test, your baseline conversion rate and the smallest lift worth finding: the calculator returns the days to run, rounded up to whole weeks, and warns when the test would run too long to trust.
- Rounded to whole weeks
- Same formula as the sample size calculator
- Runs locally
How long will this test take?
Duration = visitors needed in all groups ÷ visitors who enter the test per day. The sample per group matches the A/B test sample size calculator for the same inputs.
The control’s current rate for the page and event you will test. Enter 4 for 4%.
Relative: 10 means 4% → 4.4%. Absolute: 0.4 means 4% → 4.4% (percentage points).
The smallest lift worth detecting, in the unit chosen above.
Eligible visitors per day before the traffic split. 12k and 1,200 both work.
100 if every eligible visitor is in the test; 50 if you hold half of them out.
Control plus variants. 2 for a classic A/B test, 3 for A/B/C.
The false-positive rate you accept. 5% is the usual choice.
The chance of detecting the lift if it is real. 80% is common.
Recommended run (days)
28
Rounded up to whole weeks.
Days of traffic needed
27
Total sample ÷ daily test traffic, rounded up.
Visitors per group
39,475
Same as the sample size calculator.
Total visitors
78,950
Visitors per group × groups.
If you change the effect size
The same test at half, three quarters, 1.5× and twice your minimum detectable effect. Your row is highlighted.
| Relative MDE | Absolute | Per group | Days | Whole weeks |
|---|---|---|---|---|
| +5.0% | 0.20 pp | 154,304 | 103 | 15 wk |
| +7.5% | 0.30 pp | 69,379 | 47 | 7 wk |
| +10.0% | 0.40 pp | 39,475 | 27 | 4 wk |
| +15.0% | 0.60 pp | 17,943 | 12 | 2 wk |
| +20.0% | 0.80 pp | 10,317 | 7 | 1 wk |
Share these inputs: the values go after the # in the link, so they never reach a server.
https://tapcub.com/tools/ab-test-duration-calculator
Version v1.0Updated Update history
What the four numbers mean
The plan is a stopping rule: decide the end date before the test starts, and read the result once, on that date.
Recommended run
The days of traffic needed, rounded up to a multiple of seven. With the default inputs, 26.3 days of traffic becomes 4 weeks, 28 days. Start and stop on the same weekday.
Days of traffic needed
Total sample ÷ visitors who enter the test per day. Halving the traffic share doubles it: the default test at 50% share needs 53 days instead of 27.
Visitors per group
Identical to the A/B test sample size calculator: 39,475 per group for a 4% baseline and a 10% relative lift at 5% significance and 80% power.
The effect-size table
Shows how sensitive the duration is to the effect you choose. Halving the minimum detectable effect roughly quadruples the duration: 103 days instead of 27.
Whole weeks are a minimum, not a guarantee of a clean result. Holidays, campaigns and releases inside the test window can still distort it.
Plan the calendar in four steps
Fill in the traffic you really have, after targeting, not the site total.
-
1
Measure the baseline
Take the conversion rate of the exact page and event you will test over a recent, normal period.
-
2
Pick the effect worth finding
Choose relative or absolute and enter the smallest lift that would justify shipping the change.
-
3
Enter the traffic
Daily visitors who reach the page, and the share of them you put into the test. Add a group for every extra variant.
-
4
Read the plan
Book the recommended number of whole weeks. If the warning appears, use the table to see what a bigger effect would buy you.
From sample size to whole weeks
The sample per group comes from the standard two-sided, two-proportion z-test with a fixed sample and no continuity correction — the same formula as the A/B test sample size calculator, so both pages give the same number.
- An absolute effect is converted to relative first: relative = absolute ÷ baseline (0.4 points on 4% = 10%)
- Daily test traffic = daily visitors × share of traffic in the test
- Whole weeks remove the day-of-week bias: a 10-day test that starts on a Friday covers two Fridays, Saturdays and Sundays but only one of each other weekday
- The 8-week warning is a rule of thumb, not a statistical limit: past two months, cookie churn and seasonality start to blur who is in which group
Visitors per group
n = ( z₁₋α/₂ √(2p̄(1 − p̄)) + z₁₋β √(p₁(1 − p₁) + p₂(1 − p₂)) )² ÷ (p₂ − p₁)²
p₁ = 4%, p₂ = 4.4% → 39,474.8 → 39,475
Days of traffic
days = ⌈ n × groups ÷ (daily visitors × share) ⌉
78,950 ÷ (3,000 × 100%) = 26.3 → 27
Recommended run
run = 7 × ⌈ days ÷ 7 ⌉
27 days → 4 weeks → 28 days
Same test, different traffic and effect sizes
Every row uses a 5% significance level, 80% power and two groups.
| Baseline | Relative MDE | Daily test traffic | Per group | Days of traffic | Recommended run |
|---|---|---|---|---|---|
| 4% | 10% | 3,000 | 39,475 | 27 | 4 weeks |
| 4% | 10% | 1,500 (50% share) | 39,475 | 53 | 8 weeks |
| 4% | 10% | 1,000 | 39,475 | 79 | 12 weeks ⚠ |
| 4% | 5% | 3,000 | 154,304 | 103 | 15 weeks ⚠ |
| 20% | 10% | 3,000 | 6,510 | 5 | 1 week |
Rows marked ⚠ trip the 8-week warning. The 20% row needs only 5 days of traffic but still runs a full week. Calculated with this tool.
How test calendars go wrong
Most broken tests were stopped on the wrong day rather than sized wrong.
Stopping mid-week
Ending on day 10 because the sample arrived gives some weekdays twice the weight of others. Run whole weeks, start and end on the same weekday.
Stopping as soon as it looks significant
The duration assumes you read the result once, at the end. Checking every day and stopping at the first p below 0.05 inflates false positives well above 5%.
Using the site’s total traffic
Only visitors who reach the tested page and pass the targeting count. Entering site-wide visitors makes the test look shorter than it is.
Letting a test run for months
Very long tests collect noise: returning visitors on new devices land in another group, and campaigns and seasons move the baseline.
Where the inputs come from
TapCub does not split traffic or run experiments — your testing tool does that. TapCub gives you the numbers to plan with and a way to read conversion per group.
Baseline from funnels
Conversion between any two steps, with a fixed time window: the baseline for the step you will test.
Learn moreDaily traffic per page
Visitors per page and per day, with bots counted separately so the daily traffic is real people.
Learn moreBreak down by variant
Send the group as an event or user property from your page, then break funnels and trends down by it.
Learn moreA/B test duration questions
Still have a question?
Chat with the team behind TapCub — we usually reply within a few hours on business days.
How long should I run an A/B test?
Until each group reaches the sample size you planned, rounded up to whole weeks. With a 4% baseline, a 10% relative lift and 3,000 visitors a day in the test, that is 26.3 days of traffic, so 4 weeks.
Why round up to whole weeks?
Visitors behave differently on weekdays and weekends. A test that covers each weekday the same number of times keeps that pattern from tilting the result. It also means starting and stopping on the same weekday.
What is the minimum duration for an A/B test?
One full week, even when the sample arrives sooner. Many teams prefer two weeks so that a single unusual day carries less weight.
How long is too long?
There is no hard limit, but past about 8 weeks the sample gets noisier: people clear cookies or switch devices and get counted again in a different group, and seasonality moves the baseline. The calculator warns from 8 weeks.
Should I use a relative or an absolute effect?
Either, as long as you know which one you mean. A relative 10% lift on a 4% baseline is 4% → 4.4%; an absolute 1-point lift is 4% → 5%, which needs about 6,745 visitors per group instead of 39,475.
How does the traffic share change the duration?
The duration is inversely proportional to it. Putting 50% of eligible visitors into the test doubles the days; the sample per group stays the same.
Does adding a variant make the test longer?
Yes. Each group needs the full sample, so three groups need 1.5 times the traffic of two: 40 days instead of 27 in the default example.
Update history
New features, fixes and data refreshes for this tool, newest first.
-
· v1.0
LaunchedFirst release.
Missing something or found an error? Tell us — requests that come up often go into the next update. All tool updates
Before you read the result
When the test ends, first check that the groups got the split you configured, then compare the conversion rates.