Free tool
A/B test sample size & MDE calculator
Work out how many visitors and weeks a test needs, or flip it around: given the traffic you have, how big a lift must you see before you can trust the result?
You need to run it for
11 weeks
to detect a +10% relative lift (3% → 3.3%).
- Visitors per variant
- 53,210
- Total visitors
- 106,421
- Conversions per variant
- ~1,596
- Variants
- 2
Too slow to be useful. Aim for a bigger change, a higher-traffic page or a metric earlier in the funnel.
Lift versus time at this traffic
The smaller the lift you want to prove, the longer the test runs.
| Relative lift | Visitors per variant | Test length |
|---|---|---|
| +5% | 207,938 | 42 weeks |
| +10% | 53,210 | 11 weeks |
| +20% | 13,914 | 3 weeks |
| +30% | 6,454 | 10 days |
| +50% | 2,517 | 4 days |
How it works
Size the test before you run it
Waiting for significance without sizing the minimum detectable effect first is how teams run underpowered tests for months and end up with false positives or nothing.
The calculator uses the standard two-sided z-test for two proportions with an even traffic split. For a baseline rate p and a variant rate p′, the visitors needed in each variant are:
n = ( zα/2·√(2·p̄(1−p̄)) + zβ·√(p(1−p) + p′(1−p′)) )² ÷ (p′ − p)²
Here p̄ is the average of the two rates, zα/2 is 1.96 for 95% confidence and zβ is 0.84 for 80% power. The “what can I detect?” mode solves the same equation for the lift instead.
A common shortcut, δ = 2.8 × √(2p(1−p)/n), is close but optimistic for big lifts. Take 5,000 visitors per variant at a 3% baseline: the shortcut gives a minimum detectable lift of about 32%, while the full formula gives about 34%. The calculator uses the full formula.
Baseline
Your current conversion rate. Lower rates need far more traffic.
Lift
The relative improvement you want to prove. Halve it and you need about four times the visitors.
Confidence & power
Your tolerance for false positives and for missed wins. Stricter settings lengthen the test.
Low traffic?
When the MDE is bigger than your expected lift
If the calculator says you need +40% and you expect +5%, a classic A/B test is mathematically useless. These are the better options.
Test higher up
Run the change on the parent category, a high-traffic landing page or a sitewide element such as navigation or checkout, then roll a proven pattern down to the niche page.
Use an earlier metric
A rate of 8% (add to cart) needs far fewer visitors than a rate of 1% (purchase). Optimizing a nearer step is only worth it if it reliably moves revenue, so check that link first.
Make bolder changes
A high MDE means button colors and micro-copy are unmeasurable. Test structural changes: a new layout, fewer form fields, a different offer or price model.
Match the method to the risk
Sequential testing lets you stop early on a clear winner without inflating false positives. Bayesian tools express uncertainty differently, but neither makes a tiny sample informative.
Questions about A/B test sizing
What is the minimum detectable effect (MDE)?
The MDE is the smallest change in conversion rate your test can reliably tell apart from noise, given your traffic, baseline rate, confidence and power. A smaller MDE needs more visitors. If your MDE is bigger than the lift you realistically expect, the test will most likely end inconclusive.
How many visitors do I need for an A/B test?
It depends on your baseline conversion rate and the lift you want to detect. At a 3% baseline, 95% confidence and 80% power, detecting a +10% relative lift takes about 53,000 visitors per variant (about 106,000 for a two-variant test), while a +20% lift takes about 14,000 per variant. Use the calculator above for your own numbers.
What is the difference between relative and absolute lift?
Absolute lift is the change in percentage points (3.0% to 3.3% is +0.3 points). Relative lift is that change divided by the baseline (+0.3 points on 3.0% is +10%). This calculator uses relative lift, because it is how most teams state expected wins.
Why use 95% confidence and 80% power?
They are conventions. 95% confidence means a 5% chance of calling a winner that is really noise (a false positive). 80% power means a 20% chance of missing a real win (a false negative). Raising either makes the test longer. Lower-stakes tests can accept 90% confidence.
How long should an A/B test run?
Until it reaches the planned sample size, rounded up to whole weeks so every day of the week is counted equally. Do not stop early because a result looks significant: repeatedly checking a fixed-sample test inflates false positives. Sequential testing methods exist for teams that need to monitor continuously.
What if my traffic is too low to reach the required sample size?
Test on a higher-traffic page, pick a metric earlier in the funnel (add-to-cart instead of purchase), make a bolder change that could plausibly move the metric 30% or more, or accept a lower confidence level for low-risk decisions. Small tweaks on low-traffic pages cannot be measured.
Does this calculator work for more than two variants?
Yes. Choose up to five variants; traffic is split evenly and each variant is compared with the control. It does not adjust for multiple comparisons, so with several challengers treat borderline results with caution or raise the confidence level.
Want a test plan that fits your traffic?
Send me your site and I’ll record a free 2-minute teardown: where the biggest measurable wins are, and what to test first.